Aggregatore notizie

Porti & ambiente — le notizie raccolte

Aria, clima, elettrificazione, acque e biodiversità. 39532 articoli raccolti da fonti istituzionali e specializzate, classificati per area ambientale e linkati al porto di riferimento.

Articoli per area ambientale
reset
Brindisi e il Porto che aspetta: serve un'idea, non un'altra stagione di attesa - BrindisiReport
📰 BrindisiReport Media 📅 2026-08-07 📍 Bari it
Brindisi e il Porto che aspetta: serve un'idea, non un'altra stagione di attesa BrindisiReport
Brindisi e il Porto che aspetta: serve un'idea, non un'altra stagione di attesa Un'analisi pubblicata a fine luglio dal Sole 24 Ore lo mette nero su bianco: l'Italia sta spostando oltre l'Adriatico una parte crescente della propria profondità logistica e infrastrutturale Panoramica del porto di Brindisi BRINDISI - C'è un'Europa che in questi mesi ridisegna le proprie rotte commerciali, energetiche e persino militari attraverso l'Adriatico. E c'è una Brindisi che si ferma per l'estate come si è fermata troppe altre volte: con il petrolchimico che si ridimensiona, il ciclo del carbone che si avvia alla chiusura, l'indotto che trema, e una politica locale più impegnata a misurare i propri equilibri interni che a leggere ciò che si muove fuori dai confini comunali. Fuori dal Palazzo, la città aspetta ancora una risposta alla domanda che pone da anni: quale economia prenderà il posto di quella che sta finendo? Non è una domanda che si può liquidare guardando solo dentro i confini comunali. Perché lo scenario attorno a Brindisi, nel frattempo, è cambiato. E lo ha fatto più in fretta di quanto il dibattito politico locale sembri aver capito. Un'analisi pubblicata a fine luglio dal Sole 24 Ore lo mette nero su bianco: l'Italia sta spostando oltre l'Adriatico una parte crescente della propria profondità logistica e infrastrutturale, integrando porti, ferrovie, strade e reti energetiche fino al cuore dei Balcani e al Mar Nero. Il Corridoio VIII, l'asse che collega l'Adriatico all'Europa sud-orientale, non è più un dossier di sviluppo regionale tra i tanti. Con la Dichiarazione di Tirana del febbraio 2026, firmata dai ministri degli Esteri di Italia, Albania, Bulgaria, Macedonia del Nord e Romania, è diventato un'infrastruttura strategica europea a tutti gli effetti, con ricadute che vanno dal commercio alla mobilità militare atlantica. In questa mappa, la Puglia non è comparsa. L'analisi individua in Bari, Brindisi e Taranto una possibile “costellazione logistica distribuita”: tre porti non in concorrenza tra loro, ma con ruoli distinti nel Ro-Ro, nei carichi pesanti, nello stoccaggio, nel supporto navale, nella capacità di intervento in emergenza. E Brindisi ha già un titolo che pochi scali del Mezzogiorno possono vantare: nel luglio 2023 ha ospitato il primo incontro ministeriale dedicato proprio al Corridoio VIII. Un capitale politico e simbolico. Che però resta tale, sulla carta, finché non si traduce in una strategia infrastrutturale vera. Del resto, di potenzialità Brindisi non ne manca. Il porto, la logistica, l'economia del mare, l'aerospazio, la ricerca: sono asset che poche città del Mezzogiorno possono vantare tutti insieme, e che il territorio conosce a memoria, li ha citati in ogni convegno, in ogni piano, in ogni promessa elettorale degli ultimi vent'anni. Il problema non è mai stato individuare i punti di forza. È stato metterli a sistema, dargli una direzione, tradurli in cantieri prima che in slide. Ed è qui che il territorio brindisino ritrova un'idea che conosce da tempo, anche se non è mai riuscito a metterla a terra: un polo logistico intermodale mare-ferro, capace di collegare in modo strutturato il traffico del porto con la rete ferroviaria nazionale e con i corridoi europei. Se ne discute, a fasi alterne, da oltre vent'anni. Oggi lo scenario descritto dal Sole 24 Ore lo rende semplicemente meno rinviabile. Un porto che vuole essere nodo del Corridoio VIII, e non un terminale periferico, deve poter muovere merci pesanti e carichi prioritari senza penalizzare il traffico civile. Deve poter contare su una ferrovia integrata davvero. E deve reggere una competizione che non dorme: la Cina promuove da anni, con la Belt and Road Initiative, la propria idea di connettività verso i porti balcanici e adriatici; la Russia, dal canto suo, tratta i corridoi di trasporto come sistemi di governance prima ancora che come infrastrutture fisiche. Ma la partita dei porti, oggi, non si gioca più solo su banchine, gru e binari. Si gioca sempre di più sul terreno tecnologico. Le stesse esigenze che il quadro europeo del Corridoio VIII porta alla luce sistemi digitali affidabili, interoperabilità tra reti diverse, protezione dei nodi critici dai rischi cyber, monitoraggio dei flussi in tempo reale valgono punto per punto per la logistica portuale e per le catene di approvvigionamento che vi transitano. Un'infrastruttura senza un investimento parallelo in digitalizzazione, sicurezza informatica e analisi avanzata dei dati di filiera, per dirla con le stesse categorie usate per le opere fisiche incompiute, non è capacità strategica. È un investimento a metà. Non si tratta di un dettaglio tecnico da lasciare agli addetti ai lavori. È la differenza tra un porto che subisce il traffico che gli viene assegnato dall'alto e un porto che lo governa, lo programma, lo vende come servizio ad alto valore aggiunto. Le città che oggi attraggono investimenti logistici internazionali non lo fanno più solo mostrando banchine profonde e piazzali ampi: lo fanno dimostrando di saper garantire tracciabilità dei flussi, sicurezza dei dati, capacità predittiva sui traffici, resilienza dei sistemi. È un linguaggio nuovo, che la portualità meridionale fatica ancora a parlare. È questo l'orientamento che Brindisi dovrebbe darsi con urgenza, più di un altro cantiere, più di un altro protocollo d'intesa: non limitarsi a incassare il posizionamento geografico che la storia e la geopolitica le stanno consegnando, ma dotarsi degli strumenti — digitali, computazionali, di sicurezza delle infrastrutture critiche che trasformano quel posizionamento in valore economico reale. In imprese. In competenze. In occupazione. Un porto intelligente, prima ancora che un porto grande: è questo, e non altro, che impedisce a un territorio di restare semplice punto di transito tra un Nord che decide e un Sud che subisce. E qui il discorso torna, inevitabilmente, alle istituzioni locali. Le rassicurazioni non bastano più. Tavoli ministeriali, protocolli d'intesa, annunci di investimento producono fiducia soltanto quando si trasformano in cantieri, occupazione e prospettive concrete, non quando restano dichiarazioni buone per un titolo di giornale. Una città che chiede agli investitori di credere nel proprio futuro non può, nello stesso momento, alimentare incertezza con scelte amministrative confuse o prive di visione. La coerenza, in questa partita, vale quanto le infrastrutture fisiche: senza quella, anche il miglior progetto tecnologico resta lettera morta. Si dirà che queste sono indicazioni di metodo, non un progetto già scritto. Ed è giusto che sia così: non spetta a chi osserva il territorio da una posizione indipendente. non da progettista, non da parte in causa, disegnare al posto di chi dovrà tradurre tutto questo in atti amministrativi e in investimenti. Spetta però a chi osserva dire con chiarezza da che parte soffia il vento internazionale. Perché oggi la vera responsabilità di chi governa non è più decidere se innovare, è decidere in fretta. L'autunno che arriva dirà se Brindisi ha capito la lezione: che questo non è un rischio lontano, ma un'occasione con una data di scadenza. Il porto, la logistica, l'innovazione che li accompagna non sono un capitolo dell'agenda cittadina. Sono, oggi più che mai, l'agenda stessa. E l'agenda, a differenza delle promesse, ha sempre una scadenza. Rimani aggiornato sulle notizie dalla tua provincia iscrivendoti al nostro canale whatsapp: clicca qui Seguici gratuitamente anche sul canale Facebook: https://m.me/j/Abampv2kioahdYYR/
→ Apri originale
MP calls for urgent pollution control measures in Vizag - The Hindu
📰 The Hindu 📅 2026-08-07 en Aria · inquinamento Salute · ambiente
MP calls for urgent pollution control measures in Vizag The Hindu
Published- August 07, 2026 08:53 pm IST - VISAKHAPATNAM Visakhapatnam MP M. Sribharat submitting a representation to Union Minister Sarbananda Sonowal to address pollution in Visakhapatnam, at New Delhi on Friday. | Photo Credit: SPECIAL ARRANGMENT Visakhapatnam MP M. Sribharat met Union Minister for Ports, Shipping & Waterways, Sarbananda Sonowal, in New Delhi and submitted a detailed representation seeking stronger and time-bound measures to effectively control pollution arising from port operations in Visakhapatnam. In his representation, the MP highlighted that, given Visakhapatnam’s unique geographical setting and rapidly expanding industrial and port activities, environmental protection must receive the highest priority. He stressed that economic growth and environmental sustainability must progress hand in hand to safeguard public health and improve the quality of life of citizens. As a follow-up to the high-level review meeting convened on July 10, at the Visakhapatnam Collectorate with officials of the Andhra Pradesh Pollution Control Board (APPCB), the District Administration, Visakhapatnam Port Authority, Adani Gangavaram Port, major port operators, and other concerned departments, Mr. Sribharat apprised the Union Minister of the discussions and recommendations made during the meeting. The review had examined dust pollution generated during cargo loading, unloading, storage, and transportation activities, and identified a series of time-bound mitigation measures. The MP explained that Visakhapatnam’s bowl-shaped topography restricts the dispersion of airborne pollutants, causing particulate matter to remain trapped within the city. He noted that port-related activities are among the significant contributors to the increasing levels of PM10 and PM2.5. While acknowledging the initiatives already undertaken by the Visakhapatnam Port Authority and Adani Gangavaram Port, including mechanisation of cargo handling, deployment of fog cannons, automated sprinkler systems, HDPE cargo covering, truck tyre washing facilities, and development of green buffer zones, he observed that public concerns over pollution continue to persist. The MP urged the Union Minister to expedite the implementation of the Harit Sagar – Green Port Guidelines, 2023, and to ensure complete mechanisation of dust-intensive bulk cargo handling operations. He also requested that all port terminals be mandated to install enclosed conveyor systems, scientific cargo storage facilities, automated dust suppression systems, and wind barriers to minimise dust emissions. The MP further proposed the formulation of a comprehensive Standard Operating Procedure (SOP) for dust-free bulk cargo handling across all major ports in the country. He recommended establishing an Environmental Accountability Framework for port authorities and cargo handling agencies, incorporating measurable environmental performance indicators (KPIs), quarterly compliance reporting, public access to real-time air quality information, annual third-party environmental audits, and stringent enforcement mechanisms for repeated violations. Published- August 07, 2026 08:53 pm IST This is a premium article, available exclusively to subscribers. The Hindu delivers independent, credible journalism that explains events with depth, context, and perspective. *This offer is only for first-time subscribers.
→ Apri originale
Tagging study reveals new insights into iconic humpback song - Oceanographic Magazine
📰 Oceanographic Magazine 📅 2026-08-07 en
Tagging study reveals new insights into iconic humpback song Oceanographic Magazine
The study maps humpback whale songs to complete dive cycles for the first time, revealing how physical strain shapes vocal displays and opening new paths for marine conservation. The complexity of a humpback whale’s iconic song is directly tied to how deep it dives, a new study suggests. The findings help solve a long-standing mystery surrounding what dictates the structure of whale songs – offering a rare glimpse into their underwater behaviour and potentially serving as a vital conservation tool in an increasingly busy ocean. The study by Julia Zeh, a PhD graduate at Syracuse University’s College of Arts and Sciences, is the first to map an entire humpback whale song structure to a complete dive cycle. Researchers attached non-invasive, tablet-sized recording tags to 16 male humpback whales off Maui, Hawaii, between 2018 and 2024 using long poles and drones. The suction-cup devices simultaneously recorded high-resolution audio and 3D movement before detaching naturally. The data revealed extraordinary physical endurance, with some whales singing continuously for up to 16 hours. Crucially, recurring musical themes consistently aligned with different phases of the dive, expanding on earlier observations that had only identified changes in songs near the surface. Humpback songs follow strict, repeating patterns (“themes”), but some sections are wildly complex while others are basic and predictable. Marine biologists couldn’t explain why the song was structured this way or what caused a male to switch themes.Connecting sound directly to depth helps solve this long-standing biological puzzle by showing how physical demands shape underwater vocal displays. Themes sung at a whale’s deepest, most stable depths show the greatest acoustic variation – meaning more complex, varied singing. Scientists think this pattern points to sexual selection, where a trait evolves not because it helps an animal survive, but because it helps attract mates. Zeh said that males appear to push their acoustic capabilities at their deepest points to stand out in competitive breeding grounds. Conversely, song elements performed during ascent and descent are far more constrained, shaped by the physical demands of managing oxygen, buoyancy and changing water pressure while vocalising. The research could also give marine conservationists a new way to monitor humpbacks using passive underwater acoustic recorders already deployed across the world’s oceans. Researchers could establish clearer baselines for normal whale behaviour and better identify when commercial shipping, sonar or increasing ocean noise may be disrupting these protected animals. “This gives us direction for which parts of the song to investigate if we’re interested in variation, which is likely driven by sexual selection,” said Dr Zeh, who conducted the research in Professor Susan Parks’ Bioacoustics and Behavioral Ecology Lab. While the physiological constraints identified in the study are likely to apply to humpback whale populations globally, the researchers note that song structure differs between populations in different ocean basins. Additional tagging studies will therefore be needed to determine whether the same patterns hold elsewhere. “Projects like this one, where we combined acoustics with movement data, gave me the tools to ask exciting questions about whale behaviour and the opportunity to be creative in how we look at well-studied phenomena like humpback whale song,” said Dr Zeh. Click here for more from the Oceanographic Newsroom "*" indicates required fields The worlds biggest ocean news.Free to your inbox.Every weekday. Enjoy so much more from Oceanographic Magazine bybecoming a subscriber.A range of subscription options are available. Enjoy so much more from Oceanographic Magazine bybecoming a subscriber.A range of subscription options are available.
→ Apri originale
Portonovo più inclusiva: il Comune sostiene il progetto per rendere la spiaggia più accessibile alle persone con disabilità - Corriere Adriatico
📰 Corriere Adriatico Media 📅 2026-08-07 📍 Ancona it
Portonovo più inclusiva: il Comune sostiene il progetto per rendere la spiaggia più accessibile alle persone con disabilità Corriere Adriatico
ANCONA - Il Comune di Ancona rafforza il proprio impegno per l'inclusione nella Baia di Portonovo, sostenendo nuove iniziative e servizi capaci di rendere il territorio sempre più accessibile. L'Amministrazione comunale sostiene il progetto dell'Associazione Scuba Inside A.S.D. per la realizzazione di attività subacquee inclusive rivolto a bambini e ragazzi, con una particolare attenzione alle persone con disabilità. Il progetto L'iniziativa trasforma la Baia di Portonovo in un laboratorio a cielo aperto di integrazione sociale e benessere. Attraverso percorsi personalizzati, il progetto offre alle persone con disabilità l'opportunità di vivere un'esperienza unica in mare, trasformando il contatto con l'ambiente marino in un'occasione di libertà, scoperta, autonomia e inclusione. Un opportunità per l'accessibilità "Sosteniamo con grande convinzione questo progetto perché offre alle persone fragili un'importante opportunità di sperimentare nuove esperienze e partecipazione” dichiara l'Assessore alle Politiche Sociali Manuela Caucci. "Portonovo non è soltanto una delle perle naturali del nostro territorio, ma rappresenta l’opportunità per creare percorsi accessibili. Offrire a tutti ed in particolare alle persone con disabilità, la possibilità di vivere il mare in sicurezza grazie a professionisti specializzati significa realizzare concretamente il diritto di tutti a partecipare, sperimentare e crescere in contesti sicuri e accoglienti. Si tratta di un'iniziativa che traduce in azioni concrete il nostro impegno per una comunità sempre più inclusiva, accessibile e attenta alle esigenze di tutti i cittadini.
→ Apri originale
Retiran de Gran Tarajal el pesquero que llegó con 20 migrantes - Noticias Fuerteventura
📰 Noticias Fuerteventura 📅 2026-08-07 es
Retiran de Gran Tarajal el pesquero que llegó con 20 migrantes Noticias Fuerteventura
→ Apri originale
MEPC 85: Japan tables softer GFI targets and direct contributions - engine.online
📰 engine.online 📅 2026-08-07 en Clima · decarbonizzazione
MEPC 85: Japan tables softer GFI targets and direct contributions engine.online
Japan has proposed easing greenhouse gas fuel intensity (GFI) reduction targets from 2030 onwards and replacing the IMO Net-Zero Fund with a "direct contribution" mechanism. The Net-Zero Framework (NZF) was approved at MEPC 83. It introduced a two-tier GHG fuel intensity (GFI) standard requiring ships to progressively reduce well-to-wake greenhouse gas (GHG) emissions from 2028 onwards, benchmarked against a 2008 baseline of 93.30 grams of CO2-equivalent per megajoule (gCO2e/MJ). The GFI reduction targets comprise a base target and a direct compliance target, ranging from 4-17% in 2028 and tightening to 30-43% by 2035. Japan's submission to the IMO's Marine Environment Protection Committee (MEPC 85) proposes easing the reduction trajectory from 2030 onwards while keeping the 2028-2029 targets unchanged. Japan says the revised draft amendments "do not represent the preferences of Japan" and are intended as a basis for discussion aimed at reaching a convergence acceptable to all member states. The base target for 2030 should be relaxed from 85.84 gCO2e/MJ to 86.8 gCO2e/MJ. In 2035, it should follow a less aggressive path to 72.9 gCO2e/MJ, instead of 65.31 gCO2e/MJ. Japan proposes an adjustment to the direct compliance target, easing it from 73.71 gCO2e/MJ to 74.6 gCO2e/MJ in 2030. For 2035, it should be 60.7 gCO2e/MJ instead of 53.18 gCO2e/MJ. As an alternative, Japan sets out a flat direct compliance target of 72.9 gCO2e/MJ – a 21.9% reduction – for every year from 2028 to 2035. That would be stricter than the current draft's targets through 2030, and looser from 2031 onwards. It proposes removing the 65% base target for 2040, leaving the 2036-2040 reduction factors to be set by the Committee by 1 January 2032, "taking into account the global availability, affordability and scalability of new fuels and technologies". In the approved NZF, ships that miss the direct compliance target pay $100/mtCO2e in Tier 1 remedial units, while those that also miss the base target pay $380/mtCO2e in Tier 2 remedial units. Ships that overcomply can earn surplus units to trade or bank for future use. Japan has suggested removing the opportunity to purchase remedial units entirely. Ships that fail to meet required GFI targets can purchase surplus units from over-complying ships, or they can make contributions directly to eligible projects of their own choosing, rather than paying into the IMO Net-Zero Fund. The mechanism largely mirrors the categories of eligible projects in the NZF's revenue distribution provisions, replacing rewards for zero and near-zero fuels with support for the uptake of new fuels and technologies. It also sets out a governance framework under which flag administrations would nominate eligible projects, the IMO would determine which projects qualify, and the proposing administration would audit the entity implementing the project and report the results back. The direct contributions would be calculated using the same pricing as the remedial units in the current draft, at a minimum price of $100/mtCO2e for Tier 1 emissions and $380/mtCO2e for Tier 2 emissions. Japan's proposal has dropped regulation 39 in the approved NZF, which sets the threshold for zero and near-zero GHG emission technologies and fuels ZNZs and provides rewards from the IMO Net-Zero Fund for ships using them. Japan builds on the proposal it first tabled at MEPC 84. However, under IMO procedures, draft amendments to MARPOL are normally required to be circulated at least six months before they can be adopted. This means a proposal submitted after June would not ordinarily be eligible for adoption at the resumed second extraordinary session (MEPC/ES.2) on 4 December. In this case, member states could choose to waive that requirement and proceed with adopting the NZF at MEPC/ES.2. Alternatively, they could delay adoption to a later session, such as in 2027, said Tristan Smith, co-lead of the Shipping and Oceans Research Group at University College London. By Konica Bhatt Please get in touch with comments or additional info to news@engine.online
→ Apri originale
Oltre 47 milioni per combattere l’«inverno demografico» lucano - La Gazzetta del Mezzogiorno
📰 La Gazzetta del Mezzogiorno Media 📅 2026-08-07 it
Oltre 47 milioni per combattere l’«inverno demografico» lucano La Gazzetta del Mezzogiorno
Ridare nuova vita a territori ed edifici che, ora, sono abbandonati, per restituirle alla popolazione generando benessere economico e sociale. È una sfida che guarda alla rinascita dei comuni lucani, in particolare quelli afflitti da spopolamento ed “inverno demografico” quella che la Regione Basilicata, ed in particolare il Dipartimento ambiente, vogliono portare avanti con il programma di rigenerazione urbana, presentato ieri, ufficialmente, ad amministratori e parti sociali. Oltre 47 milioni di euro sono i fondi pronti ad approdare nei paesi lucani per ridisegnare piazze, recuperare edifici abbandonati e restituire luoghi di aggregazione alle famiglie, ai giovani e agli anziani. “Non è la solita posa della prima pietra né la semplice asfaltatura di una strada isolata” fanno sapere dalla Regione. Piuttosto, è una nuova stagione di interventi con meno opere isolate ma una vera e propria strategia per restituire qualità, servizi ed opportunità per i territori. I dettagli del “Piano di interventi rivolti alla valorizzazione ambientale e paesaggistica degli insediamenti urbani” ossia il cosiddetto “Bando Borghi” sono stati al centro di un dibattito destinato ai primi cittadini che vogliono candidare i loro progetti. Sul piatto ci sono 47 milioni 420mila e 379,61 euro derivanti dalle risorse Fsc 2021-2027, destinati ai Comuni e alle Unioni di Comuni. Già perché per rendere più capillare il programma di rigenerazione urbana è deciso che ogni singolo progetto potrà beneficiare di un finanziamento compreso tra il milione ed i 2 milioni e mezzo di euro, in grado di coprire fino al 100% dei costi ammissibili. “Si tratta di una misura che prevede rigenerazione urbana, attività di recupero di spazi, di luoghi ed edifici che possono essere restituiti alla comunità con la finalità di restituire coesione sociale e la centralità dei piccoli e grandi comuni della Basilicata. Una misura rivolta alle Amministrazioni comunali. SI favorisce la rete ed è ovviamente una misura che favorisce la transizione energetica e la sostenibilità nelle pre-declinazioni: ambientale, sociale ed economica” commenta l’assessore regionale all’Ambiente, Laura Mongiello, evidenziando come “rigenerare significa prima di tutto prendersi cura delle persone e dei luoghi che abitano”. “Non vogliamo finanziare opere pubbliche slegate dal contesto, ma costruire una strategia unitaria che metta insieme architettura, ambiente, servizi e inclusione sociale. Vogliamo restituire una funzione viva agli spazi che, nel tempo, l’hanno persa, combattendo concretamente lo spopolamento dei nostri borghi e rendendo più vivibili e sostenibili le nostre città” aggiunge ancora l’assessore. Tra le principali novità c’è il superamento della logica delle graduatorie rigide. Le proposte ritenute idonee, infatti, “non scivoleranno in una lista d’attesa passiva, ma accederanno a una vera e propria procedura negoziale: un tavolo di confronto continuo e diretto tra Regione e Comuni per affinare la qualità del progetto, verificare la sostenibilità di gestione, la governance e i tempi di realizzazione”. I comuni che vorranno presentare le loro idee, potranno farlo presentando progetti organici che mettano insieme “il recupero di edifici pubblici, la valorizzazione di aree verdi, piazze e percorsi pedonali e ciclabili, orientati alla sostenibilità e all’adattamento ai cambiamenti climatici”. Obiettivo: dare parità di accesso a tutti i comuni, dal piccolo paese delle aree interne al grande centro urbano. “Nessun comune sarà penalizzato e le opportunità di partenza saranno identiche per tutti, basandosi esclusivamente sul merito progettuale e sull’ascolto dei bisogni reali delle comunità” sottolineano ancora gli esponenti regionali. “Sono progetti che devono avere una strategia, parliamo di progetti integrati di rigenerazione urbana. Gli interventi prevalentemente sono sul patrimonio esistente, sono sui centri abitanti, sono rivolti alla popolazione residente per dare vitalità e per combattere spopolamento, frammentazione socio economica e soprattutto la poca centralità che soffrono i centri urbani” conclude Angelino Mazza, funzionario della direzione generale del Dipartimento ambiente. In questo modo si potrà recuperare il patrimonio esistente, limitare il consumo di suolo ed intervenire su superfici permeabili ed aree degradate. In poche parole, rigenerare il territorio della Basilicata.
→ Apri originale
600-horsepower 2027 Dodge Charger Super Bee Launch Edition Storms the Hive as Quickest, Fastest, Most Powerful Dodge Super Bee Ever
📰 PRNewswire 📅 2026-08-07 en
AUBURN HILLS, Mich., Aug. 7, 2026 /PRNewswire/ -- New 2027 Dodge Charger Super Bee Launch Edition boosts SIXPACK High Output (H.O.) performance to a SAE-certified 600 horsepower, surpassing the previous-generation Super Bee by more than 100 horsepower to beco…
AUBURN HILLS, Mich.,Aug. 7, 2026/PRNewswire/ -- New 2027 Dodge Charger Super Bee Launch Edition boosts SIXPACK High Output (H.O.) performance to a SAE-certified 600 horsepower, surpassing the previous-generation Super Bee by more than 100 horsepower to become the most powerful Super Bee everQuickest, fastest and most powerful SIXPACK Charger, the Super Bee improves on Charger Scat Pack acceleration and quarter-mile performance with a 3.6-second 0-60 sprint and 11.8 quarter-mile timeIntroduced to the Dodge lineup in 1968 and to Charger in 1971, the SIXPACK-powered Super Bee revives one of the brand's most iconic nameplates during the Charger 60th anniversary celebration, blending track-capable performance, serious hardware and unmistakable muscle car iconographyEngine enhancements for the 600-horsepower SIXPACK H.O. include 56mm Garrett turbochargers, larger cooling system, torque reserve and retuned engine calibrationIncreased powertrain performance pairs with upgraded brakes, functional aerodynamics and extensive development to create the most track-focused Dodge Super Bee everUnique Charger Super Bee cooling and aerodynamic upgrades increase SIXPACK cooling capacity by more than 50%Dedicated Charger Super Bee hardware includes:16-inch Brembo vented rotors and six-piston front/four-piston rear fixed calipers that deliver ultimate track performance, providing three times more brake fade resistance than any SIXPACK-powered Charger and double the brake fade resistance of the previous-generation SRT Charger20-by-11-inch forged wheels wrapped in Goodyear Eagle F1 Supercar 3 305/35ZR20 tires at all four corners that improve grip, handling and overall capabilityNew dual-valve adaptive damping suspension harnesses performance for sharper cornering and greater control, both on track and streetUnique Charger Super Bee cooling and aerodynamic upgrades increase SIXPACK cooling capacity by more than 50%Charger Super Bee performance development leveraged advanced simulation tools, including Stellantis' Automotive Research and Development Centre (ARDC) in Windsor, OntarioCharger Super Bee Launch Edition includes unique colors, graphics, wheels, brake color, badging and interior appointments:Offered exclusively in three exterior colors: new-for-2027 Sucker Punch and Shady, along with Diamond BlackExterior features gloss-black or Sucker Punch bodyside "stinger" Super Bee graphics, gloss-black 20-inch wheels and Hyper Lime brake calipersNew Super Bee logo highlights the Charger's twin-turbo, SIXPACK powerNew cloth high-back performance seats feature embroidered Super Bee logo on headrests and unique Sour Lemon accent stitching, complemented by a Super Bee instrument-panel badgeCharger Super Bee Launch Edition will be offered as a limited-production model; pricing, ordering and availability details will be announced later this year Dodge reignites a legendary nameplate during the 60th anniversary of the Dodge Charger and injects peak firepower into the SIXPACK-powered Charger era: the new 2027 Dodge Charger Super Bee Launch Edition is a SAE-certified 600-horsepower evolution of the SIXPACK-powered Charger that creates the quickest, fastest and most powerful SIXPACK Charger - and the most powerful Dodge Super Bee - ever. The new 2027 Dodge Charger Super Bee Launch Edition is engineered to attack with more power, more cooling, more braking and more repeatable performance from the next generation of American muscle. The Super Bee boosts the twin-turbo SIXPACK H.O. engine with a custom engine calibration, larger 56mm Garrett turbos and enhanced thermal capacity, and further sharpens performance with serious track-focused hardware, including massive 16-inch Brembo brakes standard for the first time on the SIXPACK Charger, 20-by-11-inch wheels with 305 summer tires standard for more grip, functional aerodynamics and adaptive damping suspension. "The Super Bee name has always stood for maximum Dodge attitude: bold, accessible performance with the hardware to back it up," said Matt McAlear, Dodge CEO. "With the new 2027 Dodge Charger Super Bee Launch Edition, we're not just bringing back a legendary badge, we're honoring that legacy by outperforming every vehicle that's adorned the Super Bee badge since 1968. 600 horsepower, bigger turbos, massive Brembo brakes, serious cooling, functional aero, and dedicated track and drag technology make this the quickest, fastest and most powerful Super Bee ever, and a clear signal that Dodge performance is only getting louder, faster and more relentless." The new Dodge Charger Super Bee Launch Edition is engineered for drivers who demand more than a one-run hero number. Straight-line speed still anchors the Dodge formula, but Super Bee widens the attack with a complete performance envelope built around acceleration, braking, cooling, track consistency, high-speed stability and driver confidence. Charger Super Bee Powertrain: Heart of the Most Power Dense Muscle Car EverAt the heart of the 2027 Dodge Charger Super Bee Launch Edition is the most potent SIXPACK-powered Charger ever engineered. A Super Bee-exclusive high-output version of the 3.0-liter twin-turbo SIXPACK engine features larger 56mm Garrett twin turbochargers, resulting in 600 horsepower and 531 lb.-ft. of torque, establishing a new performance benchmark for the SIXPACK era with 200 horsepower per liter of displacement, making the Charger Super Bee Launch Edition the most power dense muscle car ever. More Boost, More Airflow, More Repeatable PowerThe Super Bee is engineered to dominate both the drag strip and the road course, with an improved 0-60 mph sprint and quarter-mile pass, while delivering measurable gains in acceleration, braking performance, lap-time capability and thermal durability. The additional performance comes from far more than a simple horsepower increase. Engineers reworked the powertrain package to create a broader, harder-hitting performance envelope that delivers stronger acceleration at every point in the rev range and maintains that performance under repeated hard use. Compared with the SIXPACK H.O. Charger Scat Pack, the Super Bee powertrain delivers more peak power, a broader torque curve, more powerful boost response, stronger top-end acceleration and greater sustained output during extended high-performance driving. The result is a Charger that pulls harder out of corners, charges more aggressively toward redline and continues making power deep into the speedometer. The larger Garrett Motion 56mm turbochargers deliver over 30 psi of boost and spool harder than the SIXPACK Charger Scat Pack, with peak torque available 11.8% earlier at 3,000 rpm and maintained all the way through 5,900 rpm. The revised turbo package is engineered to increase both airflow and efficiency, helping unlock the highest sustained output of any SIXPACK-powered Charger. The larger turbochargers push 7.5% additional airflow through the engine, feeding the SIXPACK powerplant with a greater volume of air under demanding conditions. The increased airflow helps the Super Bee maintain stronger top-end performance while supporting more consistent power delivery during repeated acceleration runs. The Charger Super Bee's personality extends beyond raw output. Revised turbocharger and exhaust tuning create a more aggressive soundtrack, amplifying the SIXPACK's turbocharged character with sharper response, enhanced induction sound and more pronounced exhaust overrun under deceleration. Engineers also modified the intake system to maximize airflow. Modifications to the intake system, charge-air path and intercoolers include a charge air cooler duct thermal wrap, helping the Super Bee breathe more efficiently to sustain performance under the most demanding conditions. To withstand the extreme demands of producing 600 horsepower, Dodge strengthened key components throughout the powertrain. Reinforced, larger half-shafts that are nearly 10% stronger support higher output, additional boost, increased temperatures and track-duty durability. Powertrain calibration plays an equally vital role in the Super Bee's performance advantage. Exclusive software tuning sharpens throttle response, accelerates boost buildup, optimizes power delivery, refines Launch Control operation and enhances Track Mode capability. Additional thermal management and repeatability calibrations help ensure the Super Bee delivers consistent performance run after run, lap after lap. The result is a powertrain engineered to do far more than generate a bigger number on paper. Every component, calibration and airflow improvement was developed to help the Charger Super Bee accelerate harder, respond faster, pull stronger at higher speeds and maintain peak performance longer than any previous SIXPACK Charger. The Super Bee Launch Edition elevates the performance capability of the SIXPACK-powered Charger lineup in every measurable category and rockets past every previous Dodge Super Bee, including the last-generation HEMI®-powered model, improving on the 2023 Charger Super Bee's 4.5-second 0-60 mph time with a 3.6-second sprint - 20% quicker - and improving quarter-mile ET from 12.4 seconds to 11.8 seconds shaves six-tenths of a second off the quarter-mile pull. METRIC 2015 HELLCAT 2023 SUPERBEE 2026 SCATPACK 2027 SUPER BEELAUNCH EDITION IMPROVEMENT (vs. 2023 SuperBee) Horsepower 707 485 550 600 +115 Power Density (hp per liter) 114 76 183 200 +124 Torque (lb.-ft.) 650 475 531 531 +56 0-60 mph 3.7 4.5 3.9 3.6 +0.9 seconds 1/4-mile (ET) 11 12.4 12.2 11.8 +0.6 seconds Coolest Super Bee Ever: Track-ready Thermal Management Cooling SystemThe new Dodge Charger Super Bee Launch Edition doesn't just generate more power than any previous SIXPACK Charger, it's engineered to survive and sustain the heat that comes with a standard track session. Creating the quickest, fastest and most powerful SIXPACK Charger ever required a purpose-built thermal management system capable of maintaining peak performance through repeated dragstrip launches, extended road-course sessions and punishing high-temperature track conditions. The Super Bee increases SIXPACK cooling capacity by more than 50% in the high-temperature radiator circuit for engine coolant and more than doubles cooling capacity in the low-temperature radiator circuit. Those gains dramatically increase thermal headroom, allowing the Super Bee to continue attacking lap after lap in conditions that place extraordinary demands on the Charger's cooling systems. A redesigned high-flow front fascia, unique to the Super Bee, serves as the foundation of the cooling system. Developed first and foremost around performance requirements, the new fascia dramatically increases airflow by 30% to the main stack while also improving downforce by over 100 pounds at 150 mph. Larger openings, revised airflow paths and dedicated cooling channels direct significantly more air where it is needed most, helping control temperatures across the entire vehicle. To support the demands of a more powerful engine and more aggressive track capability, Super Bee expands and enhances the SIXPACK H.O.'s architecture with a series of significant cooling upgrades. The most notable addition is a set of auxiliary radiators integrated into the outboard corners of the front fascia. Surrounding components are redesigned and repositioned throughout the front of the vehicle to create room for additional cooling hardware and airflow ducting, maximizing the effectiveness of the expanded cooling package while preserving overall system efficiency. The Charger Super Bee's cooling upgrades extend beyond the radiator package. Enhanced engine cooling, charge-air cooling, transmission cooling and HVAC thermal management systems improve on-track performance. Heat exchangers, radiators, intercoolers, oil coolers, fans, airflow ducting and underbody airflow management are optimized as part of a comprehensive thermal management strategy developed specifically for the highest-output SIXPACK-powered Charger ever built. The Super Bee's track-tuned, extended splitter and dedicated cooling ducts work in concert with the expanded radiator package to manage airflow throughout the Charger. Auxiliary cooling channels and brake-cooling ducts aggressively feed fresh air to critical systems while helping optimize airflow around the front of the Charger. Together, the cooling hardware, aerodynamic surfaces and airflow-management strategy create a package designed to support relentless performance under the most demanding conditions. With engine coolant, charge-air, engine oil, transmission, differential and brake temperatures all requiring careful management, every element of the Super Bee cooling system was engineered with repeatability as the primary objective. By aggressively controlling heat buildup and maximizing airflow throughout the vehicle, the Charger Super Bee delivers greater durability, stronger thermal performance and more consistent output run after run and lap after lap, making it the coolest-running and most track-capable SIXPACK Charger ever built. Super Bee Packs Serious Track Hardware BrakesThe Super Bee's serious brake hardware delivers real-word results on the track, providing three times more brake fade resistance than any other SIXPACK-powered Charger thanks to a 27% larger brake pad surface area. The Charger Super Bee attacks braking zones with the same intensity it delivers on acceleration. Engineered to match the increased output of its high-performance SIXPACK powertrain, the Super Bee packs serious stopping hardware designed to dominate high-speed driving, punish track sessions and deliver relentless confidence lap after lap. 16-inch Brembo vented rotors front and rear, equipped on a SIXPACK Charger for the first time, anchor all four corners, while six-piston front and four-piston rear fixed calipers clamp down with race-inspired force. The result is immediate bite, aggressive deceleration and the fade-resistant performance drivers need when every corner demands commitment. The Super Bee's serious brake hardware delivers real-word results on the track, providing three times more brake fade resistance than any other SIXPACK-powered Charger thanks to a 27% larger brake pad surface area. The performance-focused brake hardware sharpens pedal response and increases driver feedback, creating a more direct connection between driver input and braking action. Whether charging into a tight corner or scrubbing speed, the Charger Super Bee delivers precise, confident braking that encourages drivers to push deeper into every braking zone. Stopping power means nothing if it can't be repeated. The Super Bee's braking system is engineered to withstand extreme thermal loads generated during sustained high-performance driving. Dedicated brake-cooling ducts aggressively feed fresh air to critical brake components, while the functional high-flow front fascia and optimized ducting strategy work together to evacuate heat and maintain consistent performance under pressure. The upgraded braking package was engineered to support repeated high-speed stops and demanding track use, helping reduce brake temperatures, resist fade and preserve pedal feel when conditions become most challenging. Super Bee delivers enhanced stopping performance, greater thermal capacity and stronger repeatability during aggressive driving than any vehicle in the SIXPACK Charger lineup. Wheels/TiresThe Dodge Charger features the widest body in the industry, and Super Bee backs that size with serious rubber. 20-by-11-inch forged wheels wrapped in Goodyear Eagle F1 Supercar 3 305/35ZR20 tires at all four corners put more performance on the pavement, maximizing grip, sharpening turn-in response and elevating overall handling capability. The wider stance and aggressive wheel-and-tire package perfectly complement Charger's commanding proportions, delivering a look as intimidating as its performance. The standard summer tires further amplify cornering confidence and high-speed stability. The oversized footprint helps Super Bee translate its power into relentless traction, whether attacking corners or during aggressive launches. The tire construction and tread pattern were also developed specifically for Super Bee, enhancing wet-weather performance without sacrificing the grip expected from a modern muscle car. SuspensionCharger Super Bee Launch Edition attacks every corner with a suspension package engineered to exploit the full capability of its performance hardware. At the heart of the system are new Continuous Damping Control (CDC) dual-valve adaptive dampers that deliver significantly more damping force than previous-generation systems, allowing Super Bee to corner harder, respond faster and stay flatter through aggressive maneuvers on both the street and the track. Unlike traditional adaptive systems that operate in fixed modes, the CDC system constantly adjusts damping forces in real time to maximize grip, stability and driver confidence. Super Bee's chassis hardware has been strengthened throughout to capitalize on the capabilities of the adaptive dampers. Front tension-link bushing stiffness increases by 25%, improving steering precision, while front handling-link bushing stiffness jumps by 50% to further reduce understeer and increase front-end authority. The result is quicker turn-in, more direct feedback and greater confidence when attacking challenging roads or demanding track sections. Front spring rates increase by 12% and rear spring rates by 6%, improving body control, reducing settling time and enhancing high-speed pitch balance. The rear stabilizer bar is 20% stiffer, helping Super Bee rotate more effectively through corners while reducing understeer, allowing drivers to fully exploit the immense grip generated by its 305-section performance tires. The advanced adaptive damping technology allows the Super Bee to pair with ultra-sticky summer tires while preserving ride quality during everyday driving. The suspension continuously balances comfort and performance, giving drivers the confidence of a track-focused setup without sacrificing street usability. Rebound springs integrated into the dampers keep the tires planted, reduce unwanted body motion and eliminate the floating sensation often associated with high-performance vehicles. Every turn triggers an instant response from the CDC system, which samples and adjusts damping force every 10 milliseconds (100 Hz control rate) to help manage lateral load transfer, body motion and steering response in real time. The adaptive dampers predict lateral acceleration and proactively apply damping force to the appropriate corner of the vehicle, keeping the body level and sharpening steering response. Advanced pitch, roll and heave controls optimize vehicle balance, while dedicated lateral-control strategies continuously manage inside-to-outside and front-to-rear damping distribution to deliver maximum handling precision. DrivelineThe Charger Super Bee puts more power and control directly in the driver's hands. A dedicated RWD Mode allows drivers to instantly switch from all-wheel drive to rear-wheel drive at any speed on the fly, unlocking a more aggressive, more engaging driving experience at the touch of a button. Whether attacking corners, carving through back roads or dialing up the attitude at the track, Super Bee delivers rear-drive performance on demand. Supporting that capability is a fortified driveline engineered to thrive under hard use. Super Bee features upgraded rear half-shafts developed and validated through extensive dragstrip testing, strengthening a critical connection between the powertrain and the pavement. The reinforced hardware gives the driveline the durability needed to handle repeated high-load acceleration and performance driving. Those upgrades help Super Bee capitalize on every horsepower. The strengthened rear half-shafts are designed to withstand the increased stresses generated by hard launches, aggressive throttle application and the increased grip of the larger tire package. The result is a driveline that remains composed and confident when the demands intensify. New SIXPACK Performance Features for Super BeeSuper Bee arrives ready to dominate both the road course and the drag strip, armed with an expanded suite of performance technology designed specifically for the SIXPACK era. New Track and Drag Modes are delivered for the first time on a SIXPACK-powered Charger, and a new dedicated RWD button mounted directly on the steering wheel allows drivers to switch to RWD Mode on the fly, making it easier to send 100% of available torque to the rear wheels for drifting, donuts and old-school muscle car fun. The performance-focused steering wheel also features larger paddle shifters that are easier to reach from both the top and bottom of the wheel, ensuring quick gear changes during aggressive driving. When it's time to attack a road course, new Track Mode transforms the vehicle and is designed specifically for use on an enclosed track. Unique Super Bee steering calibrations further amplify feedback and driver connection. Full-firm damper settings reduce body motion and settling time, while aggressive yet predictable lateral controls help Charger carve through corners with confidence. Tighter steering, elevated shift points, unique powertrain calibrations, and dedicated traction and stability control tuning sharpen every response, creating a focused performance machine built to excel on smooth, dry surfaces. Drag Mode shifts the mission from corner carving to straight-line domination. Designed specifically for use on an enclosed drag strip, Drag Mode optimizes launch performance and acceleration with dedicated chassis and driveline calibration. Drag Mode uses specialized controls with reduced front rebound and increased rear compression force to maximize weight transfer and traction, helping Super Bee put every available horsepower to the ground. Torque Reserve, a popular feature in the previous-generation Charger, returns in a new form to help the SIXPACK-powered Charger deliver harder, more repeatable launches by building boost before the vehicle leaves the starting line. While in Launch Control or when the driver holds the brake and applies the accelerator, the system opens the throttle to increase airflow, then strategically delays ignition timing to limit torque at the wheels while sending higher-energy exhaust gas to the turbos. This rapidly builds boost without overpowering the brakes. When the brake is released, ignition timing is restored and the engine immediately delivers the reserved torque, reducing turbo lag and providing stronger, more consistent acceleration off the line. Race Prep, new for the SIXPACK Charger lineup, assists in managing the Super Bee's underhood temperatures between runs by keeping the electric cooling fan operating after the engine is shut off. Continued airflow through the cooling module helps evacuate residual heat from the radiator, intercooler and engine compartment, supporting more consistent performance and helping prepare the Charger for the next run. Combined with Line Lock for smoky burnouts, Launch Control for maximum acceleration off the line and expanded Performance Pages that provide real-time performance information, Super Bee delivers an arsenal of track-ready technology that empowers drivers to extract every ounce of capability from the most track-focused SIXPACK-powered Charger yet. Additional enhancements to current drive modes include Auto, which delivers the most comfortable ride while still actively sharpening steering precision and handling response. Sport Mode tightens the vehicle's reactions with more aggressive lateral controls, creating quicker steering response and giving drivers the ability to initiate and maintain a drift with greater confidence. Drive Experience Recorder Debuts for SIXPACK-powered ChargerAvailable for the first time on the SIXPACK-powered Charger in the Super Bee Launch Edition, the Drive Experience Recorder (DxR) by Cosworth lets drivers capture and relive every hard launch, fast lap and adrenaline-filled moment. Designed for both drag racing and circuit driving, the fully integrated system synchronizes video, audio and vehicle-performance data into a complete record of every run. A forward-facing 1080p camera records crisp 60-frame-per-second video, while an integrated microphone captures the sounds of the drive, creating a comprehensive performance log that combines vehicle telemetry with real-world action. Drivers can manually activate recording or set automatic triggers to ensure DxR is ready when the action starts. An integrated USB port provides convenient storage for recorded sessions, and built-in playback functionality allows immediate review directly from the vehicle. For deeper post-event analysis, drivers can export recordings right to a phone or computer or maximize the full capability of the DxR system and export the data to the AliveDrive App to review synchronized video and vehicle data, identify fastest laps, compare performance and racing lines, and examine key driver inputs to better understand and improve on-track performance. AliveDrive is available on iOS and Android devices through the Apple App Store and Google Play. Super Bee Exterior: Aerodynamic Redesign Moves Air With Purpose, Features Unique Launch Edition TouchesThe Charger Super Bee Launch Edition wears its performance mission in plain sight. From the first sketch, Dodge designers worked side by side with engineers to create a package that doesn't just look more aggressive, but also works harder to enable the additional horsepower. Major exterior elements were developed to improve cooling, increase stability and support track-ready capability while giving Super Bee a menacing presence worthy of one of Dodge's most legendary names. Leading the charge is an all-new Super Bee fascia that delivers both attitude and function. The reworked front end features a unique headlamp configuration, a new Fratzog grille badge finished in Dark Neutral Gray and a refined, stretched metal grille texture designed to maximize airflow. Larger grille openings pull more air through the top, mail slot, bottom and dedicated side ducts, feeding the cooling system, front brakes and auxiliary coolers. Below it all, a massive track-tuned extended front splitter-one-third longer than the SIXPACK Scat Pack design-helps pin the nose to the pavement at speed and works in concert with aerodynamic enhancements throughout the vehicle to create a more planted, confident performance car. The aerodynamic story continues at the rear, where a larger new performance spoiler teams with the extended front splitter to improve overall balance and stability. Unique for Super Bee, 20-by-11-inch gloss-black wheels are contrasted by Hyper Lime brake calipers featuring black SRT by Brembo lettering, signaling the serious braking performance hiding underneath. At the center of every Launch Edition is a completely reimagined Super Bee logo that reflects the new twin-turbocharged, SIXPACK-powered era of Charger performance. Inspired by the spool of a turbocharger, the legendary logo receives a modern transformation that celebrates the forced-induction character of the new Super Bee while honoring its heritage. To celebrate the return of the iconic nameplate, Dodge launches Charger Super Bee with exclusive Launch Edition models that combine the performance upgrades of the package with three distinctive visual themes. The centerpiece is the new high-impact Sucker Punch exterior color, a vibrant yellow-green hue that delivers a fresh, modern interpretation of Dodge's long history of heritage yellow colors. Super Bee Sucker Punch Launch Edition vehicles pair the eye-catching paint with gloss-black bodyside "Stinger" Super Bee graphics, a gloss-black hood patch with Super Bee text and a painted gloss-back rear fascia that integrates seamlessly with the gloss-black rear graphics. The Super Bee Launch Edition is also available in Diamond Black exterior color, which layers in Sucker Punch accents throughout the vehicle, including a Sucker Punch-painted rear spoiler and rear fascia, along with matching Sucker Punch "Stinger" bodyside stripes. A gloss-black hood patch incorporates Sucker Punch stripes that extend out from the hood's SIXPACK bezel and Super Bee text for additional contrast. The third Super Bee Launch Edition option comes in the new Shady exterior color and takes a more menacing approach, pairing a subtle gray finish with the same gloss-black graphics to create a dark, stealthy, tone-on-tone appearance. By combining body color and unique design elements, each Launch Edition develops its own distinct personality while remaining unmistakably a Super Bee. Inside the Hive: Super Bee InteriorThe Charger Super Bee Launch Edition interior continues the Charger Super Bee's performance theme with unique details developed specifically for the return of the Super Bee nameplate. New seats create one of the most distinctive interior statements of the package, with Launch Edition models featuring the first cloth high-back performance seats offered in the modern Charger, combining aggressive bolstering with a unique appearance inspired by the Super Bee's track-focused attitude. The seat headrests feature embroidered versions of the new turbo-spooled Super Bee logo, while unique Citrus Peel "tilt" pattern seat inserts and contrast Silver and Sour Lemon stitching create an impactful, race-inspired look that stands apart from every other SIXPACK-powered Charger. A new leather performance steering wheel, unique to Super Bee, anchors the driver experience. The steering wheel features a dedicated RWD button, decorated with the silhouette of the Charger SIXPACK, that gives drivers immediate access to rear-wheel-drive mode. Larger performance paddle shifters provide easier access from both the top and bottom of the wheel, while a Citrus Peel 12 o'clock marker and contrasting Silver and Sour Lemon stitching reinforce the car's track-focused mission. A dedicated Super Bee instrument panel badge, featuring the Super Bee text and Bee logo, reinforces that this is no ordinary Charger. Dual-contrast stitching carries throughout the cockpit, highlighting the instrument panel, door panels, shifter boot and other key touchpoints with a blend of Silver and Sour Lemon touches that add energy and visual drama to the cabin. Testing the BeeThe Dodge Charger Super Bee was forged through an intensive validation program that blends advanced virtual engineering with relentless real-world testing. From the earliest stages of Charger development, engineers combined sophisticated simulation tools with extensive physical testing to shape a performance package engineered to deliver confidence and capability at the limit. A key part of that process is Stellantis' state-of-the-art Automotive Research and Development Centre (ARDC) in Windsor, Ontario. Working with the University of Windsor, engineers leveraged the cutting-edge Windsor driving simulator to accelerate development of both the new Charger platform and the Super Bee. Simulator work supports hard-point confirmation, chassis tuning, tire validation, steering-system development, ride-comfort assessments, aero-map definition, Vmax evaluation and focused lap-time development, while allowing engineers to conduct rapid back-to-back comparisons with minimal driver-variable noise. The advanced simulation environment enables teams to fine-tune steering feel, suspension calibrations, tire performance and aerodynamic balance long before production-intent vehicles reach the track. The environment allows the development team to instantly evaluate setup changes, optimize vehicle responses and refine performance targets in real time, helping deliver a Charger that feels precise, predictable and confidence-inspiring, whether attacking a road course or carving through a winding back road. Virtual development was only the beginning. Engineers validated the Charger and Charger Super Bee across more than six tracks throughout North America, including high-load braking events, high-temperature track sessions and grueling durability testing. Live track sessions then confirmed the results generated in simulation, ensuring the Super Bee delivers repeatable performance, sharper handling, stronger braking and the durability expected of a Dodge performance machine. Relaunching an Iconic Nameplate: 2027 Dodge Charger Super Bee Launch EditionAs Dodge celebrates the60th anniversary of the Charger, the brand marks the milestone by unleashing the quickest, fastest and most powerful Super Bee ever built. From its earliest days, Super Bee has represented Dodge performance at its most unapologetic with bold graphics, standout attitude and serious horsepower. The new Charger Super Bee carries that same rebellious spirit into the SIXPACK era, combining twin-turbocharged power, track-focused capability and unmistakable Dodge attitude to create a modern muscle car worthy of the legendary badge. The Super Bee story began in 1968 when Dodge introduced the original Coronet-based Super Bee, packing up to 425 horsepower and delivering blue-collar muscle-car performance that punched well above its weight. In 1971, the Super Bee joined the Charger lineup, becoming one of the most powerful Chargers of the original muscle-car era. The name returned in 2007 as a limited-production Charger Super Bee that revived the formula of aggressive styling, distinctive graphics and SRT performance. Most recently, the "Last Call" special-edition2023 Charger Super Beecapped the previous generation with 485 horsepower, setting a new benchmark for the nameplate - at the time. Now, the legend takes its biggest step forward yet. Built on the SIXPACK-powered Charger platform, the 2027 Charger Super Bee Launch Edition pushes the formula further with more power, more capability and a sharper performance edge than any Super Bee before it. 60 years after the Charger first hit the street, the most powerful Super Bee ever built arrives ready to write the next chapter of Dodge muscle-car history. The BeehiveThe 2027 Dodge Charger Super Bee Launch Edition will be built at theWindsor Assembly Plant. Pricing, ordering and availability details will be announced later this year. Key Specs: 2027 Dodge Charger Super Bee Launch EditionEngine:Twin-turbo, inline-six Super Bee H.O SIXPACK engineHorsepower:600Torque:531 lb.-ft.Transmission:Eight-speed automatic Torqueflite 880REDrive system:Standard all-wheel drive, on-demand 100% rear-wheel-drive modeFuel:PremiumBrembo Brakes:16-inch vented rotors, six-piston front/four-piston rear fixed calipersTires:Goodyear Eagle F1 Supercar 3 305/35ZR20Wheels:20-by-11-inch forgedCooling upgrades:Auxiliary radiators, charge air, transmission, HVACAero upgrades:High-flow new fascia with dedicated brake and cooling ducts, track-tuned front splitter and rear spoilerLaunch Edition colors:Sucker Punch, Shady, Diamond BlackProduction volume:Limited-run DodgeFor 112 years, the Dodge brand has carried on the spirit of brothers John and Horace Dodge. Today, that legacy roars louder than ever in the lineup of Dodge, America's performance brand. The Dodge Charger multi-energy lineup features: Every Charger comes standard with all-wheel drive and offers two-door coupe or four-door sedan configurations - because with performance comes choice. The Dodge lineup is also fueled by the most powerful gas SUV ever, the 710-horsepower Dodge Durango SRT Hellcat, powered by the legendary supercharged HEMI V-8 engine. The Durango SRT Hellcat Jailbreak breaks free from convention with the three-row SUV, unlocking more than 14 million potential customization combinations. The Dodge Durango R/T 392 delivers the most horsepower per dollar for a gas SUV, and the 360-horsepower 5.7-liter Durango GT HEMI AWD remains the most affordable AWD V-8 in the industry. The purchase of a SIXPACK-powered Charger Scat Pack, Charger Daytona Scat Pack or Durango SRT Hellcat model includes one day of performance driving instruction at Radford Racing School, the official Dodge//SRT high-performance driving school. Follow Dodge and company news and video on:Media website:media.stellantisnorthamerica.comDodge brand:dodge.comDirect Connection:DCPerformance.comFacebook:facebook.com/dodgeInstagram:@dodgeofficialX:@dodgeand@StellantisNAYouTube:youtube.com/dodge,youtube.com/StellantisNA For more information, please visit the Stellantis media site for North America athttps://media.stellantisnorthamerica.com. SOURCE Stellantis
→ Apri originale
$30 billion LNG project nears FID: Saipem confirms preliminary award shortly after McDermott
📰 Offshore Energy Media 📅 2026-08-07 en Clima · decarbonizzazione
Following a project update from its joint venture partner, McDermott, Italy’s engineering, drilling, and construction services giant Saipem has corroborated the progress made in moving forward with the next phase of one of Africa’s most closely watched liquefied natural gas (LNG) developments, as new activity signals continued momentum toward a final investment decision (FID) for Mozambique’s multibillion-dollar LNG project. The post $30 billion LNG project nears FID: Saipem confirms preliminary award shortly after McDermott appeared first on Offshore Energy .
Following a project update from its joint venture partner, McDermott, Italy’s engineering, drilling, and construction services giant Saipem has corroborated the progress made in moving forward with the next phase of one of Africa’s most closely watched liquefied natural gas (LNG) developments, as new activity signals continued momentum toward a final investment decision (FID) for Mozambique’s multibillion-dollar LNG project. On the heels of McDermott’sannouncement, Saipem has confirmed the signing of a letter of intent (LOI) for the ExxonMobil-operatedRovuma LNGPhase 1 project in Mozambique. The final award to the Italian giant within theSMDCjoint venture, which entails McDermott, Saipem, Daewoo Engineering & Construction, and China Petroleum Engineering & Construction Corporation (CPECC), is subject to FID and government and regulatory approvals expected within 2026. The LOI was penned with ExxonMobil, on behalf of the Area 4 partners, includingMozambique Rovuma Venture (MRV), comprising ExxonMobil,Eni, andChina National Petroleum Corporation (CNPC), as the operator of the Area 4 block with a 70% interest in the concession, andKOGAS,Empresa Nacional de Hidrocarbonetos (ENH), andADNOC’s XRG, each holding a 10% stake. The Rovuma LNG project is estimated to cost around $30 billion. The letter of intent covers limited preliminary engineering and procurement activities, aimed at advancing project definition and execution planning for an initial value of $32 million. The final award of the engineering, procurement and construction (EPC) contract to the SMDC joint venture is subject to positive FID by the Area 4 deepwater block partners and receipt of the necessary government and regulatory approvals, which is expected within 2026. This follows the front-end engineering design (FEED) contract previously handed out to the SMDC partnership in August 2024. The Rovuma LNG project involves the development of an onshore natural gas liquefaction plant on the Afungi Peninsula, encompassing two LNG trains with 12 modular liquefaction modules and an expected overall production capacity of approximately 18.6 million tonnes of LNG per year. This content is available after accepting the cookies. ExxonMobil books McDermott for Mozambique’s multibillion-dollar LNG project As part of the joint venture, Saipem intends to leverage its extensive experience in the development and execution of large-scale LNG projects and complex energy infrastructures, contributing to the engineering and procurement activities required to support the definition and preparation of the project. “This letter of intent further strengthens Saipem’s position in the liquefied natural gas sector and highlights the group’s expertise in executing large-scale, complex energy projects, contributing to the potential development of one of the most significant LNG initiatives currently under planning worldwide,”highlighted the Italian giant. The Area 4 block in the Rovuma Basin, which is estimated to contain approximately 85 trillion cubic feet of natural gas resources, encapsulates Eni’s operationalCoral Sul FLNGdevelopment, the plannedCoral North FLNGproject, and the ExxonMobil-ledRovuma LNGonshore liquefaction project. While MRV operates the Area 4 concession, ExxonMobil Mozambique is the delegated operator responsible for the construction and operation of the Rovuma LNG facilities, with start-up anticipated in 2031. Take the spotlight and anchor your brand in the heart of the offshore world! Join us for a bigger impact and amplify your presence at the core hub of the offshore energy community!
→ Apri originale
Sembra solo un porto di passaggio, ma è il borgo più colorato e cristallino della Campania
📰 Thewom.it 📅 2026-08-07 📍 Napoli it
Il volto quotidiano di Marina Grande Marina Grande è il punto in cui si concentrano traghetti, aliscafi, tour in barca, taxi, funicolare e autobus: qui passa chi arriva da Napoli, Sorrento, Ischia e da altri porti della Campania, ma anche chi si muove tra Cap…
L e prime cose che si notano a Marina Grande non sono i traghetti o la folla, ma i colori: le facciate in tinte pastello, le barche con la chiglia appena verniciata, il verde dell’acqua che cambia tono man mano che ci si allontana dalla riva. È il “portone d’ingresso” di Capri, certo, ma è anche un vero borgo marinaro dove vale la pena rallentare prima di correre verso la Piazzetta. Preferisci ascoltare il riassunto audio? Nell'articolo Il volto quotidiano di Marina Grande Marina Grande è il punto in cui si concentrano traghetti, aliscafi, tour in barca, taxi, funicolare e autobus: qui passa chi arriva da Napoli, Sorrento, Ischia e da altri porti della Campania, ma anche chi si muove tra Capri e Anacapri. Detto così sembra solo un nodo di scambio, in realtà è il quartiere più “vissuto” dell’isola. Lungo la banchina si alternano minimarket, farmacia, edicola, bancomat, bar e negozietti che vendono creme solari, gonfiabili, cappelli di paglia. Sono servizi molto utili per chi arriva in giornata e vuole organizzare al volo la propria visita, magari lasciando il bagaglio in uno dei depositi gestiti dai privati che si trovano alle spalle del porto. Alla fine del molo si incontrano la Capitaneria di Porto e l’ufficio informazioni turistiche: dietro questi edifici si concentrano le biglietterie delle compagnie marittime, dei bus e la stazione della funicolare che sale alla Piazzetta in pochi minuti. È il punto giusto per orientarsi, controllare orari, capire che tipo di giornata si vuole vivere tra mare, centro storico e tour in barca. La sera, quando le comitive rientrano verso la terraferma, il ritmo cambia del tutto: i rumori metallici delle manovre in porto lasciano spazio alle voci basse dei ristoranti di pesce, ai bicchieri che si toccano, a qualche pescatore che rassetta ancora le attrezzature vicino ai piccoli magazzini sul mare. Mare, bagni e giri in barca: cosa farai qui Chi vuole vedere Capri dal mare ha il suo punto di partenza naturale a Marina Grande. Dal porticciolo, lungo via Cristoforo Colombo, salpano i giri dell’isola, le escursioni alla Grotta Azzurra e i tour privati su gozzi o barche più grandi. Il giro completo della costa dura in genere da una a due ore, mentre i tour personalizzati possono occupare mezza giornata o tutta la giornata, con soste per fare il bagno sotto le scogliere. Per la Grotta Azzurra esistono due strade: scegliere un tour in barca che include la sosta all’ingresso della grotta, oppure combinare autobus per Anacapri e piccole barche locali. In ogni caso, l’accesso vero e proprio avviene sempre con minuscole barche a remi gestite da un operatore diverso, con biglietto separato, e le visite vengono interrotte in caso di mare mosso. Per ridurre l’attesa ha senso evitare le ore centrali della giornata. Il mare, però, inizia già dal porto. Poco oltre i punti di attracco si trova la spiaggia di Marina Grande, una distesa di ciottoli chiari che riflettono la luce e rendono l’acqua ancora più trasparente. È la spiaggia più ampia dell’isola, con un tratto libero e due stabilimenti attrezzati, Le Ondine e Lo Smeraldo, con lettini, ombrelloni, docce e ristorazione. Nonostante la vicinanza alle banchine, l’acqua è generalmente pulita e limpida, grazie al continuo ricambio marino, e il fondale che scende dolcemente la rende pratica per famiglie con bambini piccoli. Un piccolo rito che racconta bene il carattere del posto è il panino caprese di Aldo: pane croccante, mozzarella morbida, pomodoro maturo, basilico e olio extravergine. È lo street food “ufficiale” di Marina Grande, perfetto da addentare tra un tuffo e una partenza. Chi cerca un ambiente più raccolto può salire su un gozzo che, in pochi minuti di navigazione, porta ai Bagni di Tiberio, in zona Palazzo a Mare. Qui una spiaggia di ciottoli, l’acqua chiarissima e la vicinanza ai resti dell’antica villa romana legata all’imperatore Augusto creano un’atmosfera diversa rispetto al porto: meno rumore di motori, più rumore di acqua sulle pietre. Il biglietto del gozzo copre di solito andata e ritorno, mentre i servizi in spiaggia (lettini, bar, ristorante) si pagano a parte. Utili le scarpe da scoglio per entrare in acqua senza fastidi. Per chi si chiede se sia sicuro nuotare vicino a una zona così trafficata, l’area destinata alla balneazione è segnata da boe e separata dalle rotte delle imbarcazioni. Conviene comunque non superare i limiti di sicurezza e tenere d’occhio i bambini, soprattutto nelle giornate di maggiore affollamento. Tra chiese antiche, vicoli e Scala Fenicia Marina Grande A pochi minuti dal porto, alle spalle della cortina di case fronte mare, si nasconde la parte più storica di Marina Grande. Qui sorge una delle chiese più antiche di Capri, intitolata a Maria Santissima della Libera ma legata nel culto a San Costanzo, patrono dell’isola. L’edificio è interamente bianco, con una facciata lineare e un campanile dal gusto arabo, un esempio chiaro di come Capri abbia assorbito influenze diverse nel corso dei secoli. Le origini della chiesa risalgono probabilmente al V secolo, ed è uno dei rari luoghi dove si percepisce il passato religioso e comunitario dell’isola, oltre la vetrina turistica. Non lontano dalla chiesa si incontra uno degli accessi alla Scala Fenicia, il lungo collegamento pedonale tra il porto e l’odierna Anacapri. Il nome può trarre in inganno: non furono i Fenici a tracciarla, ma i Greci, tra VII e VI secolo a.C. I gradini in pietra sono 921, ancora percorribili, spesso al sole pieno. La salita richiede un minimo di allenamento, ma regala, passo dopo passo, viste sempre più ampie sul porto, sulle barche ormeggiate e sulla linea scura delle scogliere. Curiosamente, per chi ha articolazioni delicate, la discesa può risultare più impegnativa della salita. Per affrontarla con buon senso è meglio scegliere le ore più fresche della giornata, indossare scarpe chiuse con suola adatta e portare una buona scorta d’acqua. Non è un trekking estremo, ma nemmeno una passeggiata da fare in infradito tra un traghetto e l’altro. Tra un monumento e l’altro, Marina Grande è anche il luogo in cui vedere ancora qualche traccia del suo passato di paese di pescatori: barche a remi tirate a secco in inverno, magazzini bassi dove si ripongono reti e attrezzi, uomini intenti a sistemare legni e vernici. Chi ama il cinema riconoscerà una casa con lunga scala esterna che ha ospitato Sophia Loren e Clark Gable durante le riprese del film “La Baia di Napoli”, uno dei tanti frammenti che legano Capri alla storia del cinema italiano. Come organizzare al meglio la tappa a Marina Grande Marina Grande è facile da raggiungere perché ospita l’unico porto dell’isola. Dalla terraferma ci si imbarca su traghetti o aliscafi da Napoli, Sorrento, Ischia e altre località campane. Per chi arriva in treno, l’approdo più usato è Napoli: dalla stazione è sufficiente spostarsi verso i moli turistici con taxi, bus o metro. In aereo, l’aeroporto di riferimento è Napoli-Capodichino, da cui si raggiunge il porto con navette o taxi, per poi proseguire via mare. Una volta sbarcati, ci si orienta velocemente: di fronte al molo degli arrivi si trova la funicolare per Capri, che in pochi minuti porta direttamente alla Piazzetta. A breve distanza ci sono anche i bus per Capri e per Anacapri, utili se l’hotel o l’appartamento si trovano fuori dal raggio della funicolare. I taxi cabrio, tipici dell’isola, sono una scelta comoda quando si viaggia con valigie ingombranti o bambini piccoli, tenendo presente che nelle ore centrali dell’estate può esserci un po’ di attesa. Per chi visita Capri in giornata, Marina Grande è spesso solo un passaggio di un’ora per comprare biglietti, salire sul primo mezzo utile e poi rientrare. In realtà, con un minimo di pianificazione, questa zona può diventare il luogo in cui organizzare l’intera esperienza: al mattino presto il porto è più tranquillo, l’acqua è particolarmente limpida e c’è tempo per un bagno o un caffè vista mare prima di salire in centro. Conviene rientrare al porto con anticipo rispetto all’orario del traghetto del ritorno, per evitare corse affannate tra biglietterie, imbarchi e folle improvvise. Per chi viaggia con bambini, la compattezza di Marina Grande è un vantaggio: spiaggia, servizi e mezzi pubblici sono concentrati nello stesso fazzoletto di territorio. L’unica vera accortezza è nei momenti di arrivo simultaneo di più navi, quando la banchina si riempie: meglio tenere i più piccoli vicini e lontani dal bordo dell’acqua. Dal punto di vista dei pernottamenti, scegliere un albergo o un affittacamere a Marina Grande può essere una mossa intelligente se si hanno uno o due giorni a disposizione e si vuole ridurre al minimo i tempi morti tra traghetti, spostamenti e tour in barca. Si dorme in un quartiere calmo la sera, con il suono delle onde più presente del chiacchiericcio notturno, ma si deve mettere in conto di usare spesso funicolare o autobus per raggiungere Capri e Anacapri. Sul fronte accessibilità, la zona del porto è tra le più agevoli per chi ha mobilità ridotta, con banchine pianeggianti e alcune rampe. La funicolare e alcuni autobus hanno spazi dedicati, ma le spiagge, essendo in ciottoli, possono richiedere assistenza e un minimo di adattamento. Prima di prenotare è utile verificare direttamente con l’hotel e con gli operatori dei trasporti le soluzioni disponibili. Riproduzione riservata
→ Apri originale
El PSOE exige a Sanidad aclarar públicamente las gestiones en Playa Jardín tras los episodios de contaminación - elDiario.es
📰 elDiario.es 📅 2026-08-07 es Salute · ambiente
El PSOE exige a Sanidad aclarar públicamente las gestiones en Playa Jardín tras los episodios de contaminación elDiario.es
El PSOE de Puerto de la Cruz insta al Gobierno de Canarias y, específicamente, a laConsejería de Sanidad, responsable deSalud Pública, a una comparecencia para aclarar la gestión realizada tras el episodio de contaminación enPlaya Jardín por labacteria E.coli. El secretario general del PSOE de Puerto de la Cruz, Marco González, en un comunicado considera, a su juicio, que a estas alturas es “imprescindible” que las administraciones competentes ofrezcan una explicación “completa, rigurosa y transparente” sobre los hechos ocurridosentre Salud Pública y el Ayuntamiento tras los análisis realizados el pasado 22 de junioy el posterior aviso del cierre de la playa el 23 de junio que, asegura, “desoyó” el Consistorio. El PSOE recuerda que, según el Sistema de Información Nacional de Aguas de Baño Náyade, gestionado por elMinisterio de Sanidad, Salud Pública detectó un episodio de contaminación y comunicó dicha circunstancia por teléfono y por correo electrónico al Ayuntamiento en la mañana del día 23, tal y como estipula el protocolo oficial. Sin embargo, para el Partido Socialista, las explicaciones ofrecidasposteriormente por el Consistorio “han generado dudasy confusión”, especialmente después de las distintas versiones conocidas públicamente sobre la actuación seguida en aquellos días. “La ciudadanía tiene derecho a saber qué ocurrió exactamente, qué información se trasladó al Ayuntamiento, cuándo se produjo esa comunicación y por qué no se adoptaron antes las medidas necesarias para garantizar la seguridad de las personas usuarias de Playa Jardín”, señala Marco González, que reclama que la Consejería de Sanidad explique públicamente el contenido y el alcance de las actuaciones realizadas por Salud Pública, así como la aplicación del protocolo oficial en este caso. “Debido a las ocultaciones, tergiversaciones y falta de transparencia del Ayuntamiento, el responsable de la Salud Pública en Canarias tiene que explicar con claridad qué se hizo, cuándo se hizo, qué se comunicó y por qué no se procedió al cierre de la playa cuando se conocieron los resultados de los análisis”, afirma Marco González. El PSOE de Puerto de la Cruz advierte de que la falta de información “clara y las versiones cambiantes ofrecidas por el Ayuntamiento han provocado una pérdida de confianza entre la ciudadanía” respecto al estado de las aguas de Playa Jardín y a la gestión realizada por las instituciones responsables. “El problema no es únicamente la contaminación detectada, sino también la falta de transparencia posterior. Cuando las administraciones no explican con claridad lo sucedido, se genera incertidumbre y se perjudica la imagen de Playa Jardín y del propio municipio”, subrayan desde el PSOE portuense. Desde el PSOE de Puerto de la Cruz se califica de “nefasto” el espectáculo “de tintes franquistas”protagonizado por el alcalde y el primer teniente de alcalde, emulando la imagen de Manuel Fraga bañándose en Palomares en 1966. “Hay que recordar a estos representantes públicos que vivimos en una democracia consolidada, con instituciones que velan por la salud pública mediante protocolos rigurosos y análisis serios realizados por funcionarios técnicos y profesionales preparados para ejercer esta labor”, señalan desde el PSOE. Para los socialistas portuenses, este tipo de actuaciones no contribuye a despejar las dudas existentes ni a recuperar la confianza de la ciudadanía, sino todo lo contrario: “Están haciendo el ridículo y dejando perpleja a la ciudadanía, que espera de sus representantes públicos responsabilidad, seriedad y explicaciones, no una escenificación que banaliza un asunto relacionado con la salud pública”, añaden. El PSOE de Puerto de la Cruz considera que la seguridad de las aguas no puede acreditarse mediante gestos propagandísticos en bañador, sino a través de informes técnicos, análisis contrastados, transparencia institucional y explicaciones públicas claras. El PSOE denuncia que, tras solicitar hace más de dos semanas acceso al expediente municipal, aún el documento “no ha sido puesto a su disposición”. “La transparencia debe ser una obligación, especialmente cuando hablamos de salud pública y de la seguridad de la ciudadanía. No entendemos que se anuncie públicamente el acceso a un expediente y que, una semana después, ese compromiso siga sin cumplirse”, denuncia Marco González. Por todo ello, y ante el silencio sepulcral del Ayuntamiento, el PSOE de Puerto de la Cruz reclama al Gobierno de Canarias y a la Consejería de Sanidad, responsable de Salud Pública, que ofrezcan “con urgencia” una explicación detallada sobre los resultados de los análisis realizados el 22 de junio, la comunicación efectuada por Salud Pública al Ayuntamiento, las actuaciones desarrolladas por cada administración, las razones por las que Playa Jardín no fue cerrada el 23 de junio, la aplicación del protocolo sanitario correspondiente a este tipo de incidencias y las medidas adoptadas para garantizar la seguridad de las aguas y recuperar la confianza ciudadana.
→ Apri originale
Turismo dell’assenza per ritrovare spazio - La Gazzetta del Mezzogiorno
📰 La Gazzetta del Mezzogiorno Media 📅 2026-08-07 it Salute · ambiente
Turismo dell’assenza per ritrovare spazio La Gazzetta del Mezzogiorno
Forse il viaggio di cui abbiamo più bisogno oggi non è quello che aggiunge esperienze, immagini e destinazioni alla nostra vita, ma quello che ci restituisce spazio. Spazio mentale, prima di tutto. Spazio nel corpo, nel respiro, nello sguardo. Il turismo dell’assenza nasce da qui, da una domanda semplice e radicale. Che cosa accade quando scegliamo una vacanza non per riempire il tempo, ma per svuotarlo dal superfluo? Viviamo in un’epoca in cui anche il riposo rischia di diventare una prestazione. Si parte per vedere tutto, fare tutto, fotografare tutto, raccontare tutto. Eppure il corpo, la mente e il sistema nervoso chiedono spesso l’opposto. Non un’altra forma di consumo, ma una pausa reale. Assenza di rumore, di fretta, di notifiche, di stimoli continui. Presenza, invece, di natura, silenzio, lentezza, luce naturale, acqua, vento, alberi, orizzonte. Il turismo dell’assenza non è il turismo del vuoto. È il turismo di ciò che rimane quando il superfluo si ritira. Un borgo poco illuminato la sera, un sentiero nel bosco, una casa affacciata sul mare, una valle dove il cellulare prende male, un monastero, un rifugio, una spiaggia fuori stagione. In questi luoghi l’esperienza non è costruita per intrattenere continuamente il visitatore. È l’ambiente stesso a lavorare in profondità, con una grammatica antica fatta di ritmo, materia, ombra, respiro e ciclicità. Le evidenze scientifiche confermano ciò che molte persone sentono intuitivamente. Il contatto con spazi verdi e blu è associato a migliori livelli di benessere mentale, a minore stress percepito e a una maggiore qualità della vita. L’OMS riconosce il ruolo degli spazi verdi nel promuovere salute mentale e fisica, favorendo rilassamento psicologico, attività fisica, coesione sociale e riduzione dell’esposizione a caldo, rumore e inquinamento. Studi internazionali mostrano che non conta soltanto vivere vicino alla natura, ma frequentarla, attraversarla con il corpo. Camminare, sostare, respirare, guardare l’acqua o ascoltare gli uccelli sono azioni apparentemente semplici che possono diventare micro pratiche di regolazione emotiva. Da questo punto di vista, una vacanza nella natura non è solo un intervallo piacevole tra due periodi di lavoro. Può diventare una forma di medicina ambientale quotidiana, se viene scelta con consapevolezza. La ricerca sulle vacanze mostra che il riposo ha effetti positivi su salute e benessere, ma che questi benefici tendono a dissolversi rapidamente quando si rientra nella stessa vita di prima. Il punto, allora, non è solo partire. È imparare a portare a casa una parte del paesaggio. Un altro rapporto con il tempo. Una soglia più chiara tra lavoro e riposo. Una pratica regolare di contatto con la natura, anche breve, anche urbana, anche imperfetta. Il turismo dell’assenza ci educa proprio a questo. Non chiede vacanze estreme o luoghi irraggiungibili. Chiede una diversa qualità della scelta. Dormire dove si sente il buio. Mangiare senza fretta. Camminare senza auricolari. Guardare un panorama senza trasformarlo subito in contenuto. Sostare in un luogo abbastanza a lungo perché smetta di essere sfondo e torni a essere relazione. La salute non nasce solo da ciò che facciamo, ma anche da ciò a cui smettiamo di esporci. Meno rumore significa più ascolto. Meno stimoli significa più attenzione. Meno artificio significa più possibilità per il sistema nervoso di tornare a una misura biologica. In questa prospettiva la natura non è un accessorio della vacanza, ma la sua infrastruttura invisibile. È ciò che rende possibile il recupero profondo, perché parla al corpo prima ancora che alla mente. La luce naturale orienta i ritmi circadiani. Il verde favorisce il recupero attentivo. L’acqua apre lo sguardo e introduce fluidità emotiva. I paesaggi poco antropizzati riducono la sensazione di controllo continuo e ci ricordano che non tutto deve dipendere da noi. Anche l’assenza di attività programmate può diventare terapeutica, perché consente alla mente di decantare, riorganizzare pensieri, abbassare la vigilanza. Certamente non ogni luogo naturale è automaticamente rigenerativo e non ogni forma di turismo lento è davvero sostenibile. Il turismo dell’assenza richiede misura, rispetto e responsabilità. Non consuma la natura come scenario, ma la riconosce come organismo vivo. Non invade i territori, ma li attraversa con delicatezza. Non trasforma il silenzio in lusso esclusivo, ma lo considera un bene comune di salute. Per questo può diventare anche una nuova cultura dell’ospitalità, capace di valorizzare piccoli luoghi, paesaggi marginali, comunità locali, architetture essenziali, alimentazione semplice, mobilità dolce e ritmi più umani. Scegliere questo tipo di vacanza significa, in fondo, fare una dichiarazione sul modo in cui vogliamo vivere. Significa riconoscere che la salute non si costruisce soltanto negli ospedali, nelle palestre o negli studi medici, ma anche nei luoghi che abitiamo, nei paesaggi che frequentiamo, nei suoni che ascoltiamo, nella qualità dell’aria, della luce e del tempo. Il turismo dell’assenza non è nostalgia di un mondo perduto. È una risposta contemporanea all’eccesso contemporaneo. E’ il viaggio di cui oggi abbiamo tutti un po’ più bisogno, quello che non deve portarci lontano, ma farci tornare più vicini. Più vicini alla natura, al corpo, alla propria attenzione, a una forma di vita meno occupata e più viva. In un tempo che ci chiede continuamente di essere presenti ovunque, scegliere l’assenza può diventare un atto di salute. E forse anche una piccola rivoluzione culturale.
→ Apri originale
Oplontis, l'innovativa ricerca: ecco che cosa mangiavano le vittime - Il Mattino
📰 Il Mattino Media 📅 2026-08-07 it
Oplontis, l'innovativa ricerca: ecco che cosa mangiavano le vittime Il Mattino
→ Apri originale
Caldo record, segnali di rallentamento solo al Nord. Province marchigiane in 'bollino rosso' per tutto il weekend - AnconaToday
📰 AnconaToday Media 📅 2026-08-07 📍 Venezia it
Caldo record, segnali di rallentamento solo al Nord. Province marchigiane in 'bollino rosso' per tutto il weekend AnconaToday
Continua l'ondata di calore che attraversa l'Italia ormai da giorni, ma l'afa dà qualche segno di arretramento perlopiù nel Nord del Paese. Nel weekend, infatti, scende il numero delle città con il bollino rosso, che indica il livello più alto di allerta: sabato 8 e domenica 9 agosto saranno in tutto 19, contro le 26 di oggi. Secondo l'ultimo bollettino del Ministero, l'afa resterà ai massimi per: Ancona, Bari, Bologna, Campobasso, Catania, Civitavecchia, Firenze, Frosinone, Genova, Latina, Messina, Napoli, Palermo, Perugia, Pescara, Reggio Calabria, Rieti, Roma e Viterbo. A completamento del quadro delle 27 città monitorate, sabato sono attesi 8 bollini gialli (livello 1 di pre-allerta) per Bolzano, Brescia, Cagliari, Milano, Torino, Trieste, Venezia, Verona. L'unica variazione prevista per domenica è su Bolzano, che lascia il giallo e torna ad avere il bollino arancione, salendo di nuovo al livello 2 di allerta. Il bollettino delle ondate di calore emesso dal Ministero della Salute non riporta sostanziali novità per i cinque capoluoghi, che rimangono al massimo grado di allerta (bollino rosso sia sabato che domenica). In base alle previsioni, la colonnina di mercurio ad Ancona continuerà a oscillare tra i 32 e i 34 gradi, ma in altre zone delle Marche il caldo si farà sentire in misura maggiore, con i picchi nei centri dell’entroterra – in particolare Fabriano e Macerata – dove si toccheranno i 38 gradi e altre città in cui si prevedono tra sabato e domenica punte di 37° come ad Ascoli Piceno, Jesi e Fermo. ScegliAnconaTodaycome fonte preferita su Google News Vuoi navigaresenza pubblicità? Scopri come fare © Riproduzione riservata
→ Apri originale
Too Much Laundry Detergent Damages Your Duds. Here’s How Much to Use
📰 CNET 📅 2026-08-07 en
If you use too much detergent, you could be leaving yourself with stiff laundry.
It’s easy to go into autopilot on laundry day. You have your routine and the goal is to get through it with as little effort as possible. Simply dumping a capful of detergent into the tub is likely your default, but an expert explains why that, and several other careless laundry habits, may be sabotaging your wardrobe. Doinglaundrywith some precision and a few basic rules pays off with fabrics that emerge softer, look better and last longer. Getting those details right is often the difference between jeans and cotton tees that stay in rotation for years and ones that fray, shrink or fall apart too quickly. I asked a seasoned launderer and expert about best practices and common mistakes. Brahim Agzoul is housekeeping manager at Kasbah Tamadot, a five-star resort lauded for its exceptionally soft bedding, towels and linens and folded me in on the dos and don’ts on laundry day. Agzoul’s job is to ensure every load of laundry comes out clean and soft and to keep those fine linens intact for as long as possible. Here, Agzoul breaks down the biggest laundry blunders many of us make each week. You should be sorting laundry into more than just lights and darks, says Agzoul. While most people understand the importance of separating garments by color to prevent irreversible dye transfer, it is equally important to consider fabric type and washing requirements.“Mixing items can lead to damage and inefficient cleaning,” stresses Agzoul. Instead, consider a customizable, multi-compartment laundry bin that goes beyond whites, darks and dry cleaning. This will allow for more distinct subcategories, such as delicates or towels, to better identify items that may require special care or can withstand the harshness of standard machines. Additionally, be sure not to overstuff your washer; leave shirts unbuttoned and zippers open. This will ensure consistent soaking and drying, preventing snags and tears. Measure and use the right amount of detergent to avoid damaging clothing and irritating your skin. “It’s a common misconception that more detergent equals cleaner clothes,” said Agzoul. “In reality, excess soap lingers in fabrics, which could cause skin irritation.” Additionally, excess detergent, especially in high-efficiency machine models, can potentially damage clothing by seeping deeply into fibers, lingering beyond the spin mode and baking into the clothing during the drying process. For most medium loads of laundry (12 to 15 pounds), an ounce of detergent is sufficient — even less if it’s concentrated. A standard laundry cap is two ounces. Anything more, and you’re not only wasting that liquid soap, but you could be causing problems for your skin and clothes. If you’ve been using a full 2-ounce cap, try cutting it down to one for your next load. If clothes come out smelling fresh and looking clean, you can assume you’ve been using too much. Most laundry pros suggest using bleach sparingly, if at all. Most laundry professionals advise against the harshest addition of all: bleach. Instead, consider a sodium percarbonate product likeOxiClean, which is gentler on clothes and skin, yet retains the same brightening and lightening qualities as its more mainstream counterpart. And if you want your clothes softer, skip residue-causing fabric softener altogether. As an alternative, add a cup of white vinegar during the rinse cycle or a ½ cup of baking soda during the wash cycle to neutralize pH levels and ensure that detergents break down, yielding a fluffier outcome. Regularly cleaning your washing machine is the best thing you can do for it. The machine we use to clean our soiled linens can, unsurprisingly, become dirty itself, especially in nooks and crannies that trap moisture, allowing fabric-staining mold and far-from-fresh scents to develop. “Over time, residue can build up inside the machine,” explains Agzoul. “Special detergents designed for machine cleaning help prevent this.” A couple of highly reviewed options on Amazon includeActive, formulated as an easy-to-use tablet, and thismold and stain-removing gelthat works along rubber seals. The temperature at which you wash clothes matters. Recommended washing temperatures are in place not only to clean effectively but also to extend the longevity of what’s inside. As a general safety rule, especially when combining whites and darks, opt for a cold or room temperature wash to prevent fabrics from fading or, even worse, bleeding and staining other items in the drum. Keep spin speeds in mind, as well — delicate items such as bras, bathing suits and underwear with elastic can be stretched and misshapen if tossed around too rapidly. Dryer temperature is even more important for protecting your clothes over time. Not only are most Americans guilty of blasting their sensitive clothing with temperatures that rival those of a fighter jet’s engine, but we’re also putting the wrong types of clothing in the dryer. “Using too much heat or drying unsuitable fabrics in machines often leads to shrinkage,” says Agzoul, who advises everyone to look at care labels as a general rule of thumb. Additionally, it’s time to ditch the dryer sheets and replace them with wool balls. The artificially scented sheets are not only chock-full of a waxy residue that can build up in your dryer and pose a serious fire risk, but they’re also single-use and environmentally unfriendly. Always check the labels of the items before you wash them. The tags sewn inside clothing are there for a reason. We’ve broken down the meanings of each one inthis guide. Use them as a base-level guide to proper care, especially if you’re unsure where to start with washing and drying cycles and temperatures. “Care labels are often overlooked, yet they are essential. They provide guidance on washing, drying, bleaching and ironing,” says Agzoul. Of course, when all else fails, bring your favorite items to a local dry cleaner for a professional opinion. If an article of clothing holds sentimental value, it’s worth the extra expense to have it cleaned by an expert, allowing you to enjoy wearing it for years to come.
→ Apri originale
TechnipFMC highlights ‘critical’ subsea equipment delivery for $12.2B oil project
📰 Offshore Energy Media 📅 2026-08-07 en
UK-headquartered energy technology provider TechnipFMC has emphasized that strong execution momentum is keeping a giant deepwater oil project, operated by France’s energy giant TotalEnergies, in Block 58 off the coast of Suriname on track. The post TechnipFMC highlights ‘critical’ subsea equipment delivery for $12.2B oil project appeared first on Offshore Energy .
UK-headquartered energy technology provider TechnipFMC has emphasized that strong execution momentum is keeping a giant deepwater oil project, operated by France’s energy giant TotalEnergies,in Block 58 off the coast of Suriname on track. TechnipFMC, whichsecuredan integrated engineering, procurement, construction, and installation (iEPCI) contract worth over $1 billion in 2024 for theGranMorgu developmenton Block 58 offshore Suriname, has confirmed that this project continues to make strong progress. The company has reported the delivery of“critical”subsea equipment that is supporting the next phase of execution. According to the firm, the departure of the first drilling equipment and the installation of key floating production, storage, and offloading (FPSO) connection components demonstrate the impact of early collaboration, standardized engineering, and disciplined execution. “Together with TotalEnergies, we are helping bring one of Suriname’s most significant offshore developments closer to production. Congratulations to every team across our global network whose expertise and commitment continue to move GranMorgu forward,”stated TechnipFMC. This subsea equipment delivery comes shortly afterSBM Offshore & Technip Energies Joint Venture (STS JV)and TotalEnergies EP Suriname, a subsidiary of TotalEnergies,inked an agreementwithBeyond the Seato evaluate wind-assisted kite traction towing for Suriname’sGranMorgu FPSOproject. This content is available after accepting the cookies. Wind-assisted towing on the cards for SBM Offshore’s Suriname-bound FPSO Once progress was achieved on thefront-end engineering design(FEED) studies for Block 58’s two main oil discoveries,Sapakara SouthandKrabdagu, TotalEnergies, as the operator,sanctionedthe GranMorgu development located 150 kilometers off the coast of Suriname. These fields are estimated tohold recoverable reservesof over 750 million barrels. The operator initially estimated the project’s investment at around $10.5 billion, with first oil slated for 2028. However, the total investment value could reach up to $12.2 billion. With a daily capacity of 220,000 barrels of oil, the FPSO GranMorgu is designed to accommodate future tie-back opportunities to extend its production plateau. TotalEnergiesselectedSBM OffshoreandTechnip Energiesto handle the work related to GranMorgu’s FPSO, shortly after hand-picking Scotland’sADC Energyfor therig selection processto unlock recoverable reserves in water depths between 100 and 1,000 meters, whileSaipemwona $1.9 billion deal, enabling these players to work on Suriname’s first offshore oil development. Take the spotlight and anchor your brand in the heart of the offshore world! Join us for a bigger impact and amplify your presence at the core hub of the offshore energy community!
→ Apri originale
NH3 Completes $4M Placement for WAH2 Project - Fuel Cells Works
📰 Fuel Cells Works 📅 2026-08-07 en
NH3 Completes $4M Placement for WAH2 Project Fuel Cells Works
→ Apri originale
Rotterdam Now Leads the Bio-Bunkering Shift Again - ResourceWise
📰 ResourceWise 📅 2026-08-07 📍 Rotterdam en Clima · decarbonizzazione
Rotterdam Now Leads the Bio-Bunkering Shift Again ResourceWise
For much of the past two years, Singapore has been the undisputed leader in bio-bunkering, benefiting from its status as the world's busiest bunkering hub and a growing appetite for lower-carbon marine fuels. But the latest quarterly data suggests that the competitive landscape is changing. In the second quarter of 2026, Rotterdam overtook Singapore in reported bio-bunker sales for the first time since Q3 2024. This marks a significant milestone for Europe's marine fuels market. It's important to note that a single quarter does not necessarily establish a lasting trend. However, the figures point to an important shift driven by regulation, pricing incentives, and evolving fuel demand. According to aPrima CarbonZero analyst report, the Port of Rotterdam's bio-bunker sales more than doubled the total tonnage reported in the previous quarter. Bio-blended VLSFO accounted for the largest share of growth, followed by increases in bio-blended MGO, HSFO, ULSFO, and MDO. Bio-methanol and bio-blended LNG remained comparatively small segments. But both continued to demonstrate meaningful activity relative to historical levels. The increase was notable not only in absolute volume but also in market penetration. Bio-bunkers accounted for about 13% of Rotterdam's total bunker sales in Q2, up from over 6% in Q1. The change reflects both stronger demand for renewable fuels and a modest decline in conventional bunker volumes. By comparison, Singapore's estimated bio-bunker sales were less during the quarter based on reported monthly sales data, placing Rotterdam ahead for the first time in nearly two years. Rotterdam's rise coincided with a sharp slowdown in Singapore. June 2026 marked another month of declining bio-bunker activity and the lowest monthly level reported since February 2024. Every reported biofuel category weakened during the month, including bio-blended HSFO, bio-blended LSFO, and B100. The decline was even more pronounced compared with June 2025, when Singapore recorded one of its strongest monthly performances on record. No methanol or ammonia bunker sales were reported for June. The monthly figures extend a broader downward trend highlighted in the July market review. This suggests that Singapore's bio-bunkering market has lost momentum after leading global volumes through much of 2025. While demand remains an important driver, the data suggest that policy is increasingly shaping where shipowners bunker renewable fuels. The Netherlands has implemented RED III for the marine sector through a certificate-based mechanism that creates rebates for qualifying bio-bunker blends. During July, higher Dutch ZRE maritime certificate prices further increased those rebates, improving the economics of bunkering biofuels in Rotterdam. Rotterdam is currently the only major bunkering hub in the Amsterdam-Rotterdam-Antwerp (ARA) region to implement this certificate-based system. According to Burando Energies, this has helped make Rotterdam one of Europe's most cost-competitive ports for biofuel bunkering. This impact extends beyond renewable fuels. Earlier in 2026, higher fossil bunkering costs associated with Dutch renewable fuel mandates led to a shift in conventional bunker volumes toward Antwerp-Bruges. By Q2, however, Rotterdam's strong growth in bio-bunker demand helped offset declining conventional fuel sales and significantly increased the share of renewable fuels within its overall bunker market. Rotterdam was not the only European port reporting increased activity in renewable fuels. Antwerp-Bruges also recorded an increase in quarterly bio-bunker volumes, with sales rising from 20,726 tons in Q1 to 26,833 tons in Q2. Although these volumes remain well below Rotterdam's, they indicate continued expansion of biofuel use across Europe's principal bunkering hubs. The wider ARA market likewise showed a growing bio-bunker market share during the quarter. This reinforces the view that renewable marine fuels are becoming a larger component of regional bunker demand rather than remaining a niche segment. Whether Rotterdam retains its lead remains to be seen. Singapore continues to offer unmatched global connectivity and remains the world's largest overall bunkering hub. Quarterly results can also be influenced by vessel scheduling, procurement strategies and temporary market dynamics. Nevertheless, Q2 2026 highlights an increasingly important reality for marine fuel markets: competitiveness is no longer determined solely by geography or logistics. Policy frameworks, renewable fuel incentives, and carbon compliance mechanisms are increasingly shaping where biofuels are sourced and bunkered. For shipowners navigating tightening decarbonization requirements, these factors will likely become just as important as traditional considerations such as fuel availability and port infrastructure. As renewable fuel markets mature, competition between global bunkering hubs may increasingly be decided not only by who can supply the fuel but also by who can create the most appealing commercial environment for its use. ResourceWise’sBio-Bunkering at a Glancereport provides a monthly snapshot of the key market signals influencing marine biofuel adoption, including bunker sales, regional pricing, feedstock benchmarks, compliance costs, and major industry developments.
→ Apri originale
Wildfire, air quality forces cancellation of annual B.C. triathlon - Port Alberni Valley News
📰 Port Alberni Valley News 📅 2026-08-07 en Aria · inquinamento
Wildfire, air quality forces cancellation of annual B.C. triathlon Port Alberni Valley News
Published 7:00 am Friday, August 7, 2026 By Jordy Cunningham/Kelowna Capital News To protect the athletes, volunteers, and everyone involved, the upcoming annual Argus Apple Triathlon in Kelowna has been cancelled. Local wildfires and fluctuatingpoor air qualityare the reasons for the cancellation. The event was scheduled to take place on Sunday, Aug. 9. “This decision was not made lightly and is disappointing for all of us,” says Jonathan Arkle, president of the Apple Race Series Society. “We all hoped this weekend would be a successful culmination of months of hard work from our dedicated volunteers, Board of Directors, Race Director, and foremost the athletes who worked towards this goal. Those behind the scenes of the Apple Race Series are athletes themselves who are passionate about this event and what it means for the community, and we share the immense disappointment with all athletes and partners who were looking forward to Sunday’s races.” This decision to cancel was made in consultation with the event’s stakeholders, medical provider, Triathlon Canada, and Triathlon BC. Part of the decision was also not wanting to draw emergency resources away from theQuilpituk Creek wildfire. Included in the cancelled race were the Age Group National Championships for the standard and sprint distance triathlons and Aquabike races. All registered athletes will be notified in the coming days of their options and guidance related to their registrations. Triathlon Canada will be updating the impacted participants and provincial partners. “National Championships is the one event that brings our entire age group community together, but the health and safety of our athletes, coaches, and volunteers is always our number one priority,” said Rachel Macatee, Triathlon Canada interim CEO. “We can’t thank Race Director Susie Ernsting and the Apple Race Series team enough for their tireless work. The Apple Triathlon is one of the most prestigious events on our calendar, and we look forward to racing in Kelowna in the future.” Despite being cancelled this year, Arkle said the event will return in 2027. “We will pick ourselves up and regroup for the 2027 Argus Apple Triathlon, but for this year the safety of everyone involved, from our volunteers to our athletes, is our primary concern, and we appreciate the triathlon community’s understanding,” added Arkle. “Our thoughts are with those impacted by the fires, and our gratitude goes out to all those working to keep our communities safe.” The Argus Apple Triathlon has been a Kelowna staple since the 1980s.
→ Apri originale
A novel approach to UAV control: Fuzzy PID-based ANFIS (FPIDANFIS)
📰 Plos.org 📅 2026-08-07 en
This paper presents a novel hybrid control strategy, the Fuzzy PID based Adaptive Neuro-Fuzzy Inference System (FPIDANFIS), designed to enhance the trajectory tracking performance of Unmanned Aerial Vehicles (UAVs). The distinct contribution of this work lies…
This paper presents a novel hybrid control strategy, the Fuzzy PID based Adaptive Neuro-Fuzzy Inference System (FPIDANFIS), designed to enhance the trajectory tracking performance of Unmanned Aerial Vehicles (UAVs). The distinct contribution of this work lies in its data-driven design approach: historical input–output data from a baseline Fuzzy PID (FPID) controller using tracking error and its derivative as inputs and FPID control signals as outputs are leveraged to train an ANFIS model that autonomously generates the adaptive FPIDANFIS controller. This methodology differs from existing studies by enabling the new controller to inherit the baseline controller’s behavior while introducing adaptive capabilities not present in conventional FPID frameworks. A dataset of 5001 samples is generated using MATLAB’s Neuro-Fuzzy Designer toolbox for offline training with a hybrid learning algorithm and Gaussian membership functions. Once trained, FPIDANFIS dynamically tunes control actions in real time, improving robustness and adaptability under uncertainty. The controller is evaluated across three scenarios: nominal trajectory tracking, tracking with input disturbances, and tracking under parameter variations. Performance is quantified using the Integral of Time-weighted Absolute Error (ITAE). Simulation results show that FPIDANFIS outperforms the baseline FPID controller, achieving a 42% ITAE reduction under disturbances and a 22.12% reduction under parameter variations. The proposed method introduces a data-driven, adaptive extension of the FPID framework and demonstrates significant improvements in UAV trajectory tracking accuracy, making it suitable for applications requiring precise and robust maneuverability. Citation:Madebo NW, Yareshe FT, Lemma LN (2026) A novel approach to UAV control: Fuzzy PID-based ANFIS (FPIDANFIS). PLoS One 21(8): e0355158. https://doi.org/10.1371/journal.pone.0355158 Editor:Van Thanh Tien Nguyen, Industrial University of Ho Chi Minh City, VIET NAM Received:August 19, 2025;Accepted:July 16, 2026;Published:August 7, 2026 Copyright:© 2026 Madebo et al. This is an open access article distributed under the terms of theCreative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability:All data is available with in article. Funding:The author(s) received no specific funding for this work. Competing interests:The authors have declared that no competing interests exist. The adoption of Unmanned Aerial Vehicles (UAVs) has rapidly expanded across numerous civil applications, including real-time monitoring, remote sensing, search and rescue, security, precision agriculture, and infrastructure inspection. The integration of smart UAV technology has significantly transformed these applications by reducing operational risks and costs. Among various domains, civil infrastructure is projected to dominate the UAV market, contributing to a global market value exceeding 45 billion dollars [1]. Ensuring safe, reliable, and efficient UAV operations necessitates advanced automation and control systems, which encompass navigation, flight control, mission planning, and emergency handling. Despite the significant progress in UAV automation, several challenges persist, particularly in multi-UAV coordination, communication, and collision avoidance [2]. Effective flight control must ensure feasible maneuvers under adverse conditions, while navigation systems should leverage sensor fusion techniques to operate in dynamic environments, even during sensor failures. Fail-safe mechanisms, such as emergency landing protocols and return-to-home functionalities, are crucial for mitigating operational risks. Additionally, recent advancements in artificial intelligence (AI) and machine learning (ML) offer promising solutions for enhancing UAV autonomy and adaptive decision-making [3]. Quadrotors, in particular, present substantial control challenges due to their under-actuated dynamics, high sensitivity to external disturbances, and inherent instability. The complexity of motion planning in unknown and dynamic environments further amplifies the need for robust and adaptive control strategies [4,5]. Conventional Proportional-Integral-Derivative (PID) controllers, while widely used due to their simplicity, often struggle to maintain stability and performance in nonlinear and rapidly changing environments. As a result, intelligent control techniques have emerged as viable alternatives [6]. Fuzzy Logic Controllers (FLCs) have gained increasing prominence in UAV control owing to their effectiveness in managing system uncertainties and nonlinear dynamics [7–9]. Building on this foundation, the integration of fuzzy logic with neural network learning in Adaptive Neuro-Fuzzy Inference Systems (ANFIS) has enabled the design of intelligent adaptive controllers that combine heuristic reasoning with data-driven learning mechanisms [10–12]. ANFIS-based control architectures are particularly appealing for UAV applications, as they exhibit strong anti-interference capabilities, rapid convergence characteristics, and enhanced adaptability in complex and dynamic operating environments [13–15]. These strengths have motivated extensive research on ANFIS-enhanced UAV controllers, especially for trajectory tracking tasks subject to disturbances, modeling uncertainties, and parameter variations [16–22]. By leveraging both the structured reasoning of fuzzy systems and the adaptive learning capability of neural networks, ANFIS-based hybrid control methods offer a compelling pathway toward improving precision, robustness, and autonomy in aerial robotic platforms [23,24]. In summary, the existing literature underscores that hybrid control algorithms particularly those integrating fuzzy logic and learning-based adaptation play a vital role in achieving reliable and high-performance UAV trajectory tracking in the presence of disturbances and uncertainties. Recent advancements in quadrotor UAV control have increasingly focused on overcoming the challenges posed by their nonlinear, underactuated, and inherently unstable dynamics. Although proportional–integral–derivative (PID) controllers remain widely adopted because of their structural simplicity, they frequently lack the robustness and adaptability required for complex real-world conditions characterized by parameter variations and external disturbances. To address these limitations, researchers have progressively shifted toward more advanced control algorithms, with particular emphasis on adaptive fuzzy-based methods [25–29]. Fuzzy PID-based adaptive controllers have demonstrated notable improvements in robustness, adaptability, and tracking accuracy in UAV trajectory control [30–36]. Building on these advances, hybrid strategies that integrate fuzzy logic with neural networks, particularly the Adaptive Neuro-Fuzzy Inference System (ANFIS), have emerged to further strengthen adaptability and learning capability [37]. The versatility of fuzzy PID controllers has also been validated in multi-joint robotic arm systems, where their ability to manage highly coupled dynamics while ensuring stability has been emphasized [38]. In [39], an ANFIS-based task-offloading model for mobile edge computing was proposed to address dynamic and resource-constrained environments. Implemented using MATLAB’s Neuro-Fuzzy Toolbox and evaluated through a 70% training and 30% testing split, the model achieved a low error rate of 0.29505, underscoring its effectiveness. Similarly, [40] introduced an ANFIS-based autonomous UAV flight controller employing three fuzzy logic modules to regulate altitude, speed, and roll angle. The system was modeled using MATLAB and the Aerosim Aeronautical Simulation Block Set and validated through Microsoft Flight Simulator and FlightGear, illustrating its suitability for rapid nonlinear UAV modeling. For fixed-wing UAVs, [41] presented an ANFIS-based heading control technique that combined fuzzy reasoning with neural learning procedures. Implemented on an onboard microcontroller equipped with real-time navigation sensors, the method enabled fully autonomous flight. Expanding on these developments, [42] compared a genetic algorithm–optimized self-tuning PID controller with an ANFIS-based controller using the Aerosonde UAV model in Simulink/MATLAB. Simulation results revealed significantly enhanced robustness from the ANFIS controller, particularly under windy conditions. Hybrid control architectures have also gained attention. In [43], a backstepping-based PID and ANFIS hybrid controller was designed for UAV formation control in cluttered environments. MATLAB Simulink simulations confirmed its capability to maintain leader–follower formations despite obstacles. A broader review in [44] further highlighted the extensive applicability of ANFIS across engineering domains, particularly within power electronics, where it contributes to power quality improvement in multilevel converters. Beyond aerial robotics, ANFIS has been employed in automotive and energy systems. In [45], an ANFIS-based controller for anti-lock braking systems (ABS) was designed to optimize wheel slip and reduce braking distance, outperforming conventional PID and fuzzy-tuned PID controllers. Likewise, [46] developed an ANFIS-based bidirectional power management strategy for plug-in electric vehicles (PEVs), improving energy exchange efficiency between the grid and vehicle batteries while reducing harmonic distortion. Several additional studies have extended ANFIS applications to industrial control and UAV trajectory tracking. For instance, [47] implemented an ANFIS-based controller to improve accuracy in spray drying systems, resulting in increased thermal efficiency and product consistency. A trajectory tracking control strategy using an uncertainty and disturbance estimator (UDE) for quadrotors was presented in [48], while an adaptive trajectory tracking controller for quadrotor micro aerial vehicles (MAVs) was explored in [49]. To further mitigate external disturbances, adaptive disturbance compensation strategies were proposed in [50,51], followed by adaptive robust trajectory tracking methods in [52–54]. An ANFIS-based sliding mode control (ANFIS–SMC) scheme was introduced in [55] for UAV trajectory tracking and obstacle avoidance. Comparative simulations demonstrated that ANFIS–SMC significantly reduced tracking errors and improved operational safety—achieving a 0% collision probability compared to 33.33% for conventional SMC—albeit at the cost of increased computational time. More recently, ANFIS has been employed directly for UAV trajectory tracking [56], where adaptive optimization of control parameters was achieved using an input–output dataset derived from a fuzzy PID (FPID) controller. This approach improves reference tracking accuracy while ensuring adaptability to nonlinear flight dynamics. Despite these advances, achieving consistently efficient reference tracking in quadrotor UAVs using ANFIS-based controllers requires precise and systematic tuning of control parameters across diverse operating conditions. Traditional PID and Fuzzy PID (FPID) controllers often struggle to address the strong nonlinearities inherent in quadrotor systems. Motivated by these gaps, this study proposes a trajectory-tracking control strategy based on a Fuzzy PID–based ANFIS (FPIDANFIS) controller. By leveraging the input–output characteristics of an FPID controller, the proposed method adaptively tunes control parameters to deliver robust and accurate trajectory tracking under nonlinear and dynamically changing flight conditions (Table 1). https://doi.org/10.1371/journal.pone.0355158.t001 The key contributions of this work include: The proposed FPIDANFIS approach contributes to the development of intelligent UAV control frameworks, enabling enhanced maneuverability, stability, and resilience in complex environments. The development of the FPIDANFIS control system for UAV trajectory tracking follows a structured approach, comprising the following key stages: The modeling of a quadrotor system is a critical step in understanding its behavior and developing effective control strategies. Quadrotors, classified as unmanned aerial vehicles (UAVs) with four rotors, exhibit complex dynamics arising from the interaction of aerodynamics, mechanics, and control systems. This section outlines the fundamental concepts necessary for the system modeling of a quadrotor. This paper investigates the trajectory tracking control challenges associated with the ‘X’(cross) configuration quadrotor UAV. The propulsion system of this UAV consists of four motors, as illustrated inFig 1, where adjustments in rotor speeds are utilized to control both attitude and position [57]. https://doi.org/10.1371/journal.pone.0355158.g001 Proposed Model:This study adopts the ‘X’ (cross) configuration for the quadrotor model. The primary advantage of this configuration lies in its superior stability compared to the plus configuration, as the cross configuration utilizes two rotors to stabilize and guide the quadrotor’s movements [58,59]. The subsequent section focuses on Dynamics and Kinematics,of the quadrotor system. This analysis considers variables such as rotor thrust, torque, and the overall motion of the system in three-dimensional space. In the realm of the Rigid Body Control Model, we introduce how the quadrotor responds to external inputs and disturbances. This involves a comprehensive understanding of the forces and moments acting on the vehicle system. The subsequent segment encompasses the presentation of mathematical equations dictating the quadrotor’s motion. This includes translational and rotational dynamics, along with the presentation of equations of motion and their derivations from a reference frame inFig 2. To precisely depict the position and attitude of the quadrotor UAV, the Earth-fixed coordinate system is utilized as (,,,) and the body coordinate system() are introduced. https://doi.org/10.1371/journal.pone.0355158.g002 The definition of the body structure and the coordinate system is shown inFig 2. Modeling the quadrotor UAV is complex, to simplify the model establishment, the following assumptions are considered [58,59,60]. The orientation of the quadrotor is given by the rotation matrixRwhich depends on the three Euler angles (,,). Transfer Matrix() is in Inertial frame and Angular Velocityis in Body frame.To relate the two, Transfer matrix() is used as [58,59,60]: Newton-Euler Method formalismThe flight dynamics of the quadrotor UAV will be modeled using the Newton-Euler method [58,59,60,61]. The equations of motion, which combine the translational and rotational parameters which is described inTable 2of the 6DOF system (consisting of 3 position and 3 angular orientation variables), are derived based on Euler’s two laws of motion. https://doi.org/10.1371/journal.pone.0355158.t002 The angular acceleration in the body frame is expressed as: The quadrotor’s dynamics are described by four control inputs, represented as: where: The quadrotor’s translational and rotational dynamics are governed by the following equations [58,62]: The control inputs are used to calculate the rotor angular velocities, with their inverse relationship given as: Virtual control is introduced to handle the underactuated nature of a quadrotor, which has six degrees of freedom (three translational and three rotational) but only four independent actuators. Because of this limitation, the vehicle cannot directly produce lateral motion; instead, horizontal movement is achieved indirectly by tilting the aircraft to generate the required thrust components. To manage this coupling, the control system is typically divided into two layers. The outer loop (virtual control layer) computes the desired forces or accelerations required to reach a target position in the x, y, and z directions. These outputs are treated as virtual control inputs. The inner loop then converts these virtual commands into actual motor inputs by mapping them to individual rotor speeds using techniques such as control allocation or inverse kinematics. This structure simplifies the overall control problem and enables effective trajectory tracking in complex dynamics. The quadrotor is underactuated, with 4 control inputs and 6 states to be controlled. The states are indirectly controlled through the translational dynamics [58,59,60,63,62]: where,, andare the virtual control inputs. Based on these equations, the lift force (altitude control signal) and the desired Euler angles are calculated as [58]: This paper addresses the trajectory tracking control problem of a quadrotor UAV, as described inEquation (10), with the goal of ensuring precise tracking of a predefined trajectory by minimizing the tracking errore(t), defined inEquation (26). To achieve this objective, an intelligent and adaptive control strategy is required. A Gaussian membership function, given inEquation (18), is employed to enhance control adaptability, while a hybrid learning algorithm is utilized to optimize the error gradient. This optimization process generates the optimal intelligent adaptive control signalU, formulated inEquation (25). The following sections provide a detailed explanation of the ANFIS architecture and its learning mechanism, which are designed to minimize the tracking error and improve the overall trajectory tracking performance. ANFIS StructureTo illustrate the ANFIS architecture, two fuzzy IF-THEN rules based on a first-order Sugeno model are considered: where: The reasoning mechanism for this Sugeno model, which forms the basis of the ANFIS model, is illustrated inFig 3. The ANFIS architecture consists of five layers, where circles represent fixed nodes and squares indicate adaptive nodes. Each layer performs a distinct function in processing the input data. https://doi.org/10.1371/journal.pone.0355158.g003 Layer 1: Fuzzification LayerAll nodes in this layer are adaptive and compute the membership grade of the inputs using fuzzy membership functions. The outputs are given by: whereandrepresent the fuzzy membership functions associated with linguistic labels (e.g., high, low). These functions can adopt the Gaussian membership function: where,, andare parameters that define the shape of the membership function. Layer 2: Rule Firing Strength ComputationThis layer consists of fixed nodes that compute the firing strength of each rule using the AND operator: Layer 3: Normalization LayerThe nodes in this layer normalize the firing strengths from the previous layer: These normalized values,, are known as normalized firing strengths. Layer 4: Defuzzification LayerThe nodes in this layer are adaptive, computing the contribution of each rule to the final output: whererepresents the normalized firing strengths, and,, andare the consequent parameters. Layer 5: Output LayerThis layer consists of a single fixed node that performs the summation of all incoming signals to compute the final output: The ANFIS learning process employs a hybrid algorithm combining the least squares method and gradient descent to optimize both premise and consequent parameters. The training process consists of two passes: The hybrid learning approach improves convergence speed by reducing the search space dimensions compared to conventional backpropagation methods [64]. The overall output of the ANFIS model can be expressed as: which, when expanded, is given by: Rewriting in terms of individual components: The least squares method is employed to determine optimal values for the linear consequent parameters. When the premise parameters are not fixed, the search space increases, leading to slower convergence. The hybrid approach mitigates this issue by separately optimizing premise and consequent parameters. The effectiveness of the hybrid algorithm has been demonstrated in improving ANFIS training efficiency by balancing parameter optimization through complementary learning strategies [65,64]. This section details the design of Fuzzy PID (FPID) and Fuzzy PID-based Adaptive Neuro-Fuzzy Inference System (FPIDANFIS) controllers. The FPID controller combines the conventional PID structure with fuzzy logic to enable real-time adaptive gain adjustment, improving performance in nonlinear systems. Conversely, the FPIDANFIS controller employs a data-driven approach, learning from system behavior to optimize control actions for complex systems such as quadrotors.Gaussian membership functions (MFs) are widely used in fuzzy logic systems and Adaptive Neuro-Fuzzy Inference Systems (ANFIS) due to their smooth curves, differentiable nature, and their ability to closely represent real-world continuous processes. Although simpler shapes such as triangular and trapezoidal MFs require less computation, Gaussian functions are preferred when higher accuracy, smooth transitions, and strong adaptability in modeling are required. To enhance trajectory tracking, the control deviation at each instant is correlated with the system’s historical state. The generalized control scheme block diagram is shown inFig 4. https://doi.org/10.1371/journal.pone.0355158.g004 The error and its derivative are defined as: wheredenotes the desired reference signal, andrepresents the output states of the quadrotor. The FPID controller uses a fuzzy system for online PID gain tuning [58,66,67,68]. This is essential due to the nonlinear and underactuated nature of quadrotors, making classical PID control inadequate. The design involves the following steps: The structure of the FPID controller is shown inFig 5and its input/output surface plot is depicted inFig 6. https://doi.org/10.1371/journal.pone.0355158.g005 https://doi.org/10.1371/journal.pone.0355158.g006 The fuzzy inference system uses a set of linguistic rules to determine the parameters,, andautomatically [58,69,70,68]. Input variables use seven fuzzy labels (NB: Negative Big, NM: Negative Medium, NS: Negative Small, ZO: Zero, PS: Positive Small, PM: Positive Medium, PB: Positive Big), while output variables use seven labels (VVS: Very Very Small, VS: Very Small, S: Small, M: Medium, B: Big, VB: Very Big, VVB: Very Very Big). These are detailed inTables 3and4. https://doi.org/10.1371/journal.pone.0355158.t003 https://doi.org/10.1371/journal.pone.0355158.t004 For position and attitude control, the error is normalized to, the error rate to, and the output ranges are [0.2, 0.7], [0.001, 0.01], and [0.1, 0.15] for proportional, integral, and derivative gains, respectively. The system employs product-sum inference and center-of-gravity defuzzification [58,66,68,71,72]. The fuzzy controller output is: Adaptive Neuro-Fuzzy Inference Systems (ANFIS) integrate the learning capabilities of neural networks with the imprecision-handling strengths of fuzzy logic. This hybrid approach effectively models complex nonlinear relationships and adapts to uncertainties within data, making ANFIS a robust tool for outcome estimation and system control. A major limitation of fuzzy or neuro-fuzzy systems is their failure to deal with high-dimensional datasets [73]. The Adaptive Neuro-Fuzzy Inference System (ANFIS) serves as a classifier by combining fuzzy logic with neural networks in a hybrid approach. It implements a fuzzy inference system (FIS) within an adaptive fuzzy neural network framework, leveraging the explicit knowledge representation of FIS and the learning capabilities of artificial neural networks [74].Adaptive Neuro-Fuzzy Inference System (ANFIS) in prediction using data sets and fusion of fuzzy logic and neural network with a hybrid learning algorithm can be very useful in any research problem beyond engineering research even in social science research like Psychological research [75]. The Adaptive Neuro-Fuzzy Inference System (ANFIS) controller effectively addresses control challenges in uncertain systems by combining the fuzzy system’s ability to handle uncertainty with the self-learning capability of neural networks for parameter optimization [76]. For nonlinear and uncertain ship motion, an ANFIS-based course-keeping controller is designed to simplify parameter adjustment and enhance control performance [77].The Adaptive Neuro-Fuzzy Inference System (ANFIS) utilizes error and its derivative as inputs, along with the output of the fuzzy PID (FPID) controller. For this study, a dataset comprising 5001 samples, including two inputs (error and its derivative) and one output (FPID controller), is generated in the MATLAB workspace. Offline training is performed using the Neuro-Fuzzy Designer toolbox, employing a hybrid learning algorithm to optimize the system, and the gaussian membership function is used and the output is constant. The model structure is inFig 7and the proposed control strategy is depicted inFig 8whereas, the training data interface, ANFIS training using Neuro-Fuzzy Designer toolbox and its input output surface plots illustrated inFigs 9,10, and11respectively whereas, the fuzzy rule view and surface plot of the FPIDANFIS controller is dipicted inFigs 12and13respectively. https://doi.org/10.1371/journal.pone.0355158.g007 https://doi.org/10.1371/journal.pone.0355158.g008 https://doi.org/10.1371/journal.pone.0355158.g009 https://doi.org/10.1371/journal.pone.0355158.g010 https://doi.org/10.1371/journal.pone.0355158.g011 https://doi.org/10.1371/journal.pone.0355158.g012 https://doi.org/10.1371/journal.pone.0355158.g013 Table 5presents the ANFIS training configuration and parameter settings employed in the MATLAB environment. The table specifies the number and categories of membership functions assigned to the input variables, which are crucial for achieving accurate system representation. It also summarizes the training parameters, including the maximum number of epochs, the optimization algorithm adopted (e.g., hybrid learning or backpropagation), and the learning rate settings used throughout the training process to enhance convergence and model performance. https://doi.org/10.1371/journal.pone.0355158.t005 The quadrotor simulation incorporates the parameters listed inTable 6within the MATLAB environment. The performance of the proposed FPIDANFIS controller is evaluated for trajectory tracking and compared to the standard FPID controller under varying conditions. Simulations are performed using MATLAB 2023a Simulink, and the results demonstrate the superior effectiveness of the FPIDANFIS controller in achieving accurate helical trajectory tracking.The reference signal for helical trajectory tracking is defined as: https://doi.org/10.1371/journal.pone.0355158.t006 The controllers’ (FPIDANFIS and FPID) tracking performance under undisturbed conditions. The results for helical trajectory tracking are shown inFigs 14–18. https://doi.org/10.1371/journal.pone.0355158.g014 https://doi.org/10.1371/journal.pone.0355158.g015 https://doi.org/10.1371/journal.pone.0355158.g016 https://doi.org/10.1371/journal.pone.0355158.g017 https://doi.org/10.1371/journal.pone.0355158.g018 TheTable 7presents a comparison of the computational time required by the FPIDANFIS and FPID controllers for helical trajectory tracking. It can be observed that the FPIDANFIS controller exhibits a lower computational time of 6.2367549 seconds compared to the FPID controller, which requires 7.46578 seconds. This indicates that the FPIDANFIS approach is computationally more efficient, achieving faster execution while maintaining the control functionality. https://doi.org/10.1371/journal.pone.0355158.t007 For infinity trajectory tracking the reference signal is: The controllers’ (FPIDANFIS and FPID) tracking performance under undisturbed conditions. The results for infinity trajectory tracking are shown inFigs 19,20,21,22and23. https://doi.org/10.1371/journal.pone.0355158.g019 https://doi.org/10.1371/journal.pone.0355158.g020 https://doi.org/10.1371/journal.pone.0355158.g021 https://doi.org/10.1371/journal.pone.0355158.g022 https://doi.org/10.1371/journal.pone.0355158.g023 Algorithm 1: Pseudocode for triangular trajectory 1:ift< 10 2: 3: 4: 5:else if 6: 7: 8: 9:else if 10: 11: 12: 13:else 14: 15: 16: 17:end if The controllers’ (FPIDANFIS and FPID) tracking performance under undisturbed conditions. The results for triangular trajectory tracking are shown inFigs 24,25,26,27and28. https://doi.org/10.1371/journal.pone.0355158.g024 https://doi.org/10.1371/journal.pone.0355158.g025 https://doi.org/10.1371/journal.pone.0355158.g026 https://doi.org/10.1371/journal.pone.0355158.g027 https://doi.org/10.1371/journal.pone.0355158.g028 This section assesses the disturbance rejection capability and robustness check with the FPIDANFIS and FPID controllers under conditions of parameter variation and input disturbance. To replicate real-world scenarios, a random force representing wind disturbances was introduced as an external input, affecting thex,y, andzpositions of the system. Additionally, parameter variations were modeled by increasing the total quadrotor mass by 20% and the inertia values along thex,y, andzaxes by 10%. The applied disturbance force was set to 0.5 N in thex,y, andzdirections during specific time intervals: 5–10 seconds for thexaxis, 20–25 seconds for theyaxis, and 30–35 seconds for thezaxis. These conditions were designed to rigorously evaluate the system’s performance under external disturbances and dynamic parameter changes for triangular trajectory. The simulation results for triangular trajectory, incorporating the effects of the input disturbances are presented inFigs 29,30,31,32, and33. https://doi.org/10.1371/journal.pone.0355158.g029 https://doi.org/10.1371/journal.pone.0355158.g030 https://doi.org/10.1371/journal.pone.0355158.g031 https://doi.org/10.1371/journal.pone.0355158.g032 https://doi.org/10.1371/journal.pone.0355158.g033 The simulation results for triangular trajectory, incorporating the effects of the variation in parameter are presented inFigs 34,35,36,37and38. https://doi.org/10.1371/journal.pone.0355158.g034 https://doi.org/10.1371/journal.pone.0355158.g035 https://doi.org/10.1371/journal.pone.0355158.g036 https://doi.org/10.1371/journal.pone.0355158.g037 https://doi.org/10.1371/journal.pone.0355158.g038 This study undertakes a comprehensive comparison of two controllers— Fuzzy Proportional-Integral-Derivative (FPID) and proposed FPIDANFIS controller based on performance criteria in the tabular form described inTables 8–10in undisturbed conditions, disturbed(with input disturbance added) and variation in parameter conditions respectively. https://doi.org/10.1371/journal.pone.0355158.t008 https://doi.org/10.1371/journal.pone.0355158.t009 https://doi.org/10.1371/journal.pone.0355158.t010 The performance of the FPID and FPIDANFIS controllers is evaluated under three scenarios: undisturbed conditions, input disturbances, and parameter variations. The analysis is based on the Integral Time Absolute Error (ITAE) criteria as shown inTables 8–10. FromTable 8, it is observed that the FPIDANFIS controller consistently outperforms the FPID controller across all states in undisturbed conditions. The ITAE values are lower for FPIDANFIS in every state, indicating superior tracking performance. For example: The FPIDANFIS controller exhibits its strengths in handling complex nonlinearities and dynamic conditions more effectively than FPID. However, both controllers perform equally for thestate, which may suggest limited control effort is required for this state. In the presence of input disturbancesTable 9, FPIDANFIS again outperforms FPID, particularly in theyandzstates: These results demonstrate the robustness of FPIDANFIS under external disturbances. However, the performance differences in other states such asandare minimal, which indicates similar control effectiveness in these states under disturbances. For thestate, the ITAE remains zero for both controllers. In scenarios with parameter variationsTable 10, the FPIDANFIS controller continues to show superior performance. Despite these strengths, the improvements are marginal in some states such asand, which suggests limitations in FPIDANFIS when parameter variations are less critical. Both controllers exhibit limitations in specific scenarios: Overall, the FPIDANFIS controller demonstrates better robustness and tracking performance compared to FPID in all tested scenarios. However, its advantages are more prominent under challenging conditions, such as disturbances and parameter variations. The minimal differences in some states highlight areas for potential optimization in the FPIDANFIS design (Table 11). https://doi.org/10.1371/journal.pone.0355158.t011 While the proposed FPIDANFIS controller demonstrates improved performance, the study has the following limitations: This paper presents a novel control strategy for quadrotor UAV trajectory tracking, demonstrating enhanced accuracy and robustness across a range of operating conditions. The novelty of the proposed approach lies in leveraging historical data from a baseline FPID controller to design the new FPIDANFIS controller. Specifically, input–output data from the FPID, with the tracking error and its derivative as inputs and the FPID control signals as outputs, were used to train an ANFIS model, which then generates the adaptive FPIDANFIS controller. Comparative analysis shows that FPIDANFIS consistently outperforms the baseline FPID, effectively handling disturbances and system uncertainties. While the approach significantly improves adaptability and reliability, further refinements are needed to enhance control consistency and extend applicability to real-world UAV operations. Future work will focus on improving the controller’s adaptability and validating its performance under practical flight conditions. The authors would like to thank the Information Network Security Administration (INSA), Addis Ababa, Ethiopia, for institutional support and for providing a conducive working environment during the course of this research.
→ Apri originale
Decoding behavior with minimal and interpretable agent models
📰 Plos.org 📅 2026-08-07 en
Author summary Living organisms constantly process information to make decisions. Which are the computational processes that translate sequences of sensory cues into courses of actions? Since the internal states of an organism typically cannot be observed, it…
Understanding how living organisms process sensory information from their surroundings and translate it into decisions is a fundamental problem across biological scales – from biochemical signalling in single-cells to neural computations in animal brains. In this work, we address this challenge by introducing a method to reconstruct general decision processes directly from behavioral observations alone. Our approach is applicable to any biological agent and does not require prior knowledge of its internal mechanisms or its environment. Our agent model is defined by a recurrent dynamics over a discrete set of internal states which encode and process sensory information, and dictate which actions to execute. We validate our method on synthetic agents and demonstrate that we can exactly recover the agent’s behavior for non-trivial tasks. Then, we infer agent models from experimental data of rats performing evidence accumulation and of mice making decisions under uncertainty and in changing environments. In both cases, very few internal states suffice to reproduce the observed behavior with high accuracy. Crucially, the immediate interpretability of the inferred dynamics allows to understand the computational process underlying decision-making. Living organisms constantly process information to make decisions. Which are the computational processes that translate sequences of sensory cues into courses of actions? Since the internal states of an organism typically cannot be observed, it is difficult to answer this question. Even when measurements are possible – such as in large-scale recordings of neural activity – extracting computational principles from a high-dimensional dynamics still poses significant challenges. Here, we introduce a method to infer interpretable models of the decision-making process directly from behavioral trajectories, without making assumptions about the environment. Despite their simplicity, these agent models are expressive enough to accurately describe experimentally observed behaviors and reveal the structure of the underlying neural computational processes. Citation:Nicoletti G, Celani A (2026) Decoding behavior with minimal and interpretable agent models. PLoS Comput Biol 22(8): e1014585. https://doi.org/10.1371/journal.pcbi.1014585 Editor:Barbara Webb, The University of Edinburgh, UNITED KINGDOM OF GREAT BRITAIN AND NORTHERN IRELAND Received:February 17, 2026;Accepted:July 16, 2026;Published:August 7, 2026 Copyright:© 2026 Nicoletti, Celani. This is an open access article distributed under the terms of theCreative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability:The code to infer finite state controllers from behavioral trajectories is available athttps://github.com/giorgionicoletti/FSC-inference-MAPSO. Funding:The author(s) received no specific funding for this work. Competing interests:The authors have declared that no competing interests exist. Decisions are everywhere. Viruses [1], bacteria [2,3], single-cell eukaryotes [4–6], rodents, and mammals [7,8] must continuously make decisions to survive and thrive. Yet, understanding how biological and living systems collect, process, and translate information into actions, whether biochemically or through neural processes, remains a fundamental challenge across fields. A wide range of modelling frameworks to capture biological behavior has been proposed, from Bayesian models to artificial neural networks. However, these approaches typically make strong assumptions about the decision-making process at hand or sacrifice interpretability for predictive accuracy. In this work, we develop a general framework that allows us to infer interpretable decision-making strategies directly from behavioral observations across diverse agents and tasks. The theory of decision-making processes provides a general mathematical language to describe behavior as the repeated interaction between anagentand theenvironmentthat surrounds it. The boundary between agent and environment is the sensorimotor interface, where information is received and decisions are executed (seeFig 1). Now, suppose we have at our disposal a dataset of sequences of interactions between an agent and the environment. How can we infer from the data a “good” model of the decision process of the agent? In principle, an agent with a perfect memory could make decisions based upon the entire history of past interactions with the environment. In practice, living systems are constrained by finite computational resources. Thus, biological agents must compress such histories in theirinternal states, whose role is to encapsulate all the information that is available to the agent and that is relevant for the task at hand. The biological realization of the internal state can in general be a complex and very high-dimensional object, like the biochemical state of a cell [9], the activity patterns of neurons in a brain [10–12] or, in a more abstract setting, the state of an artificial recurrent neural network [13,14]. (a)We consider a generic agent that interacts with an environment via a sensorimotor interface, by taking elementary actionsa(e.g., moving in space or selecting a choice) and receives observationsyin response (e.g., sensory cues and rewards). The agent’s behavior is driven by an agent statem, which determines its actions through a decision process described by a probability distributionover the action space. At each step of the decision process, the internal state is updated to account for new observations through an internal computation process modeled by a probability distributiongover the space of internal states.(b)We assume that we only have access to a series of behavioral timeseries of the agent, correspondingto sequences of action and observation pairs.(c)Our goal is to infer the internal model with which the agent makes decisions as a finite state controller (FSC), where the agent’s state is replaced by a finite set of discrete internal states. Its decision process is represented by stochastic policydetermining the actiongiven the current internal state, and an internal computation modelwith which the agent process information by updating its internal state depending on the previous state, the action, and the observation. In the diagram, circular nodes represent internal states, square nodes actions, and diamond nodes observations. The thickness of the arrows between nodes is proportional to the corresponding transition probability, both for the agent policy(gray arrows) and the computation(colored arrows). (a)We consider a generic agent that interacts with an environment via a sensorimotor interface, by taking elementary actionsa(e.g., moving in space or selecting a choice) and receives observationsyin response (e.g., sensory cues and rewards). The agent’s behavior is driven by an agent statem, which determines its actions through a decision process described by a probability distributionover the action space. At each step of the decision process, the internal state is updated to account for new observations through an internal computation process modeled by a probability distributiongover the space of internal states.(b)We assume that we only have access to a series of behavioral timeseries of the agent, correspondingto sequences of action and observation pairs.(c)Our goal is to infer the internal model with which the agent makes decisions as a finite state controller (FSC), where the agent’s state is replaced by a finite set of discrete internal states. Its decision process is represented by stochastic policydetermining the actiongiven the current internal state, and an internal computation modelwith which the agent process information by updating its internal state depending on the previous state, the action, and the observation. In the diagram, circular nodes represent internal states, square nodes actions, and diamond nodes observations. The thickness of the arrows between nodes is proportional to the corresponding transition probability, both for the agent policy(gray arrows) and the computation(colored arrows). https://doi.org/10.1371/journal.pcbi.1014585.g001 Here, we employFinite State Controllers(FSCs) as minimal agent models defined by a discrete set of internal states [15–19]. Formally, FSCs represent a conceptual extension of input-output Hidden Markov Models and variants alike [20–23] with the key difference that, in an FSC, actions are naturally fed back into internal transitions, allowing internal states to maintain a memory of past decisions and use it to inform future ones. The main advantage of FSCs is twofold. First, the discrete structure of internal states can be interpreted in terms of behavioral modes, governed by internal processes whose dynamics are essentially low-dimensional. This is akin to the fact that the computations of RNNs [24–27] and networks of biological neurons [28–33] can be understood in terms of transitions between discrete metastable attractors, or even belief-like representations [34–37]. Describing an agent by an FSC bypasses the need for high-dimensional fast neural dynamics by directly expressing the decision-making process through a highly coarse-grained representation of the slower processes. Furthermore, recent efforts have shown that small RNNs are able to represent a variety of cognitive tasks, suggesting that compact models of behavioral strategies can achieve the performance of more complex models [38]. Second, if the focus is ondecodingthe decision process – i.e., understanding which computational processes are made by an agent – FSCs yield interpretable and tractable descriptions of decision strategies, making the computational processes at play explicit. These considerations lead us to ask: could an FSC with a relatively small number of internal states be a good model for the behavior of an agent engaged in a real task? To answer this question, we develop a method to infer FSC models from sequences of interactions between the agent and the environment alone. No prior information about the environment is required, nor does the inference ever attempt to model how observations are produced – the focus is entirely on the agent. We first test our method against synthetic agents performing a variety of tasks, from arithmetic computations to spatial navigation, and show that we can efficiently discover FSCs that perfectly reproduce behavior even with a limited number of behavioral trajectories. Then, we exploit our method to infer FSCs from experimental data in rats performing an evidence accumulation task and in mice during a reversal task, where they must make decisions in uncertain and changing environments. We find that inferred FSCs with few internal states not only reproduce observed behaviors with high accuracy, but also uncover the underlying computational processes driving behavior. Remarkably, in some cases, the inferred FSCs can be explicitly mapped into known decision architectures, such as exponentially smoothed counting or leaky competing accumulator models. This demonstrates that our approach provides interpretable descriptions that directly connect behavioral observations to potential neural implementations and generate testable hypotheses. Let us describe the agent’s decisions asactions– whereacould be a movement in a given direction, an interaction with an object, or a measurement of some quantity – andobservations– for instance, sensory cues of different types. In general, the actiontaken at time-steptby the agent may depend on the full history of past interactions with the environment,, through some probability distribution. Rather than keeping a record of past experiences, a realistic agent encodes the history in its internal state. The agent’s behavior can then be described in terms of the probabilityof taking an actiongiven its current internal state, and of the recurrent computation process, which updates the internal state by processing the information that has just been acquired. Both thepolicyand theinternal state update gare conditional probability distributions, so that in general the process is stochastic. The description of the decision process is completed by a choice of the initial distribution of internal states. Amodel of the agentis thus defined as the tuple, as we summarize inFig 1A. An FSC is an agent characterized by a discrete set of internal states. For the tasks considered in this work,andare also discrete sets of cardinalityand. We stress here that an FSC implicitly encodes both past observations and actions inthrough the recurrence of the internal computationg. In this sense, it generalizes the notion of belief about the state of the external environment that characterizes Bayesian agents [39]. Further, there is an exact mapping between FSCs and linearized gated recurrent units, a gating mechanism used in recurrent neural networks (RNNs) [40,41] (see Section D inS1 Appendix). This correspondence provides a bridge between the abstract representation of an FSC and the more biologically interpretable variables of an RNN, and generalizes input-output Hidden Markov Models [20–23] to allow internal states to retain information about past actions and use it to guide future decisions. Finally, we note that FSCs can always be made expressive enough to reproduce the behavior of an arbitrary agent over a time spanT, albeit at the punishing price of an exponentially large number of internal states. In this work, however, we focus on FSCs with relatively smallM, following the idea that internal states should represent how the agent compresses available information, rather than performing a complex act of mimicry. We assume that we havebehavioral trajectories, i.e., histories of interactions between agent and environment, each described by a sequence of actions and observations(seeFig 1B). Our goal is to infer a finite state controller that, at a given observation sequence, will reproduce the actions as closely as possible (Fig 1C). To this end, we seek FSC parameters that maximize the log-likelihood of the FSC actions given the observations, or equivalently solve the optimization problem where Above, the policy, the internal computationg, and the initial distribution of the internal states, which fully characterize the agent, are parametrized by softmax functions (see Materials and Methods). The optimization problem inEq. (1)does not require a model of how observations aregenerated by the environment. This is a consequence of observations appearing only as conditional dependencies inEq. (2). Thus, the environment may be arbitrarily complex, and observations may depend on the current action as well as on the entire history of past interactions. The maximization of the likelihood in (1) is, in general, a non-convex optimization problem. Thus, in order to infer the parameters that define the FSC, we developed a global optimization method, MAPSO, whose core idea is that multiple FSCs are initialized with random parameters and then evolved according to an interacting swarm dynamics. MAPSO is based on Adaptive Particle Swarm Optimization [42], where particles are FSCs, but the topology of the interactions changes with the metric distance during the optimization. After a fixed number of iterations, the FSC that achieved the largest log-likelihood at any point during training is retained (see Materials and Methods and Section E inS1 Appendixfor more details). We remark here that we do not add any regularization terms to the likelihood, such as sparsity constraints, although this is possible in principle. In this work, we infer FSCs with different numbers of internal statesMand analyze them separately, and find that, in all cases considered here, there exists a value ofMabove which FSCs start to accurately reproduce relevant behavioral observables. We note that the number of parameters of an FSC is dominated by the internal computation, growing asM2AY(see Methods). Therefore, in finite datasets, this effectively limits the number of states that can be reliably inferred, especially in scenarios whereAandYare large. Finally, we note that optimization can also be performed by a local search algorithm, e.g., gradient methods or an adapted Baum-Welch expectation maximization algorithm, augmented with random restarts. While we make no claim of computational supremacy, in our experience, MAPSO turns out to be more robust and efficient in locating global optima, at least for the cases discussed below. In Section F of theS1 Appendix, we show an explicit comparison with expectation maximization and stochastic gradient descent, showing that in some cases MAPSO outperforms them both. We first test our inference procedure on a purely synthetic agent performing an arithmetic task. The agent can take two actions, “A” or “B”. If the agent takes A, the environment returns a random digit between 0 and 4 with equal probability. Conversely, if it takes the B action, the observation is a digit between 6 and 9. The agent has a counter that is initially set to zero. At the first step, the agent takes a random action with equal probability, observes a digit, and adds it to the counter. At all subsequent times, the agent performs a parity check on the counter and takes action A if it is even and B if it is odd. Then, it receives a new digit and adds it to the counter (Fig 2A). We want to know if our inference algorithm can retrieve the decision-making process just from a dataset of sequences of actions and observations without any information about the inner workings of the environment and of the agent. (a) An agent can take two generic actions, “A” and “B”, and observe a digit. The agent adds the received digits to an internal counter that determines the following action according to a parity check (action A if it is even, B if it is odd). The first action is random. (b) Negative log-likelihood of an FSC withM= 3 internal states during training withN= 50 trajectories ofT= 100 steps each. The inferred FSC perfectly reconstructs the agent’s behavior, reaching the expected negative log-likelihood (dashed gray line). (c) Inferred FSC. Thestate is the initial one, where the action is taken at random (π(A|) ≈ 0.48, reflecting the empirical distribution). Thestate is dedicated to the action A, whereas theone always takes B. The transitions between the states are always deterministic and solely depend on the parity of the observation, as expected. (d) An agent performing a spatial navigation task in a T-maze 7 tiles in length. The agent enters the maze with a → action and receives an “up” (U) or “down” (D) observation with equal probability. It then either keeps moving to the right (60% probability) or steps up (↑) or down (↓) with a slight bias that depends on the first observation (30% and 10% probability, respectively). Once it reaches the T-junction, the agent chooses ↑ or ↓ solely based on the first observation, ignoring all the ones it received in the corridor. (e) Negative log-likelihood of FSCs with differentMduring training withN= 1000 trajectories. WithM= 5, the inferred FSC reaches the expected negative log-likelihood of the exact model (dashed gray line). (f) Thestate represents the initial condition, which transitions to one of two “corridor states”, either(first up observation) or(down observation). Once it reaches the end of the corridor, the transition→allows the FSC to correctly take ↑ , while→results in ↓. (g) A chemotactic agent in a one-dimensional concentration gradient. At each time, the agent may detect the presence of a chemical (observation •) or not detect it (), with a probability that depends on its concentration, which in turn depends on the position in space. If in the last two steps it received the same observation, the action is chosen at random between tumble (T), which reorients the agent without any displacement, and run (R), which moves the agent in its current direction. If the last observation was • (a detection) and the previous one was(no detection), then the agent runs. In the reversed case, it tumbles. (h) Negative log-likelihood of FSCs with differentMduring training withN= 250 trajectories ofT= 100 steps each. WithM= 4, the FSC reaches the expected likelihood. (i) The internal states of the FSC encode for the four combinations of past observations:∼ (yt−2,yt−1) = (×,×);∼ (×, •);∼ (•,×);∼ (•, •) and works as a shift-register memory. (a) An agent can take two generic actions, “A” and “B”, and observe a digit. The agent adds the received digits to an internal counter that determines the following action according to a parity check (action A if it is even, B if it is odd). The first action is random. (b) Negative log-likelihood of an FSC withM= 3 internal states during training withN= 50 trajectories ofT= 100 steps each. The inferred FSC perfectly reconstructs the agent’s behavior, reaching the expected negative log-likelihood (dashed gray line). (c) Inferred FSC. Thestate is the initial one, where the action is taken at random (π(A|) ≈ 0.48, reflecting the empirical distribution). Thestate is dedicated to the action A, whereas theone always takes B. The transitions between the states are always deterministic and solely depend on the parity of the observation, as expected. (d) An agent performing a spatial navigation task in a T-maze 7 tiles in length. The agent enters the maze with a → action and receives an “up” (U) or “down” (D) observation with equal probability. It then either keeps moving to the right (60% probability) or steps up (↑) or down (↓) with a slight bias that depends on the first observation (30% and 10% probability, respectively). Once it reaches the T-junction, the agent chooses ↑ or ↓ solely based on the first observation, ignoring all the ones it received in the corridor. (e) Negative log-likelihood of FSCs with differentMduring training withN= 1000 trajectories. WithM= 5, the inferred FSC reaches the expected negative log-likelihood of the exact model (dashed gray line). (f) Thestate represents the initial condition, which transitions to one of two “corridor states”, either(first up observation) or(down observation). Once it reaches the end of the corridor, the transition→allows the FSC to correctly take ↑ , while→results in ↓. (g) A chemotactic agent in a one-dimensional concentration gradient. At each time, the agent may detect the presence of a chemical (observation •) or not detect it (), with a probability that depends on its concentration, which in turn depends on the position in space. If in the last two steps it received the same observation, the action is chosen at random between tumble (T), which reorients the agent without any displacement, and run (R), which moves the agent in its current direction. If the last observation was • (a detection) and the previous one was(no detection), then the agent runs. In the reversed case, it tumbles. (h) Negative log-likelihood of FSCs with differentMduring training withN= 250 trajectories ofT= 100 steps each. WithM= 4, the FSC reaches the expected likelihood. (i) The internal states of the FSC encode for the four combinations of past observations:∼ (yt−2,yt−1) = (×,×);∼ (×, •);∼ (•,×);∼ (•, •) and works as a shift-register memory. https://doi.org/10.1371/journal.pcbi.1014585.g002 It would be natural to identify the internal states of this agent with the counter’s value, but their number would be very large if sequences are long. Are there more compact descriptions of the agent’s decision-making process? In fact, our inference algorithm discovers an exact representation of the agent’s behavior with justM= 3 internal states, valid for any sequence length. InFig 2B, we plot the negative log-likelihood during training with MAPSO fromN= 50 behavioral trajectories of the agent of lengthT= 100. We find that the inferred FSC reaches the negative log-likelihood expected from the agent, that is the conditional entropy of the empirical distribution of actions conditioned to observations, which in this case reduces to the entropy of the initial action distribution (see Section A inS1 Appendix). InFig 2C, we draw the inferred FSC topology. The first state,is the initial one where the action is taken at random. The other two internal states –and– are dedicated to taking the A and B action, respectively, and switch between them if an odd digit is observed. It is easy to check that this FSC exactly implements the agent’s decision-making rules without the need for a counter. We now consider the case of a synthetic agent performing a spatial navigation task (see, e.g., [43]) that requires the exact memorization of some observations, selectively ignoring other ones, and making associations between long-past and just-received observations (Fig 2D). The agent is stepping on a tiled T-maze and can take three actions: if it takes action → , it makes a step to the right, whereas actionsleave it at the same location (it bounces on the corridor walls). At the beginning of the task, the agent enters the maze from the left, always taking the → action, and receives an observation that can be either “up” or ”down” with equal probability. At subsequent steps, the agent either chooses action → most of the time or takes ↑ or ↓ with a slight bias towards the action that matches the observation received at the entrance step. As it travels along the corridor, the agent keeps receiving observations “up” or ”down” with equal probability until it reaches the T-junction and receives the observation “end”. At this time, it deterministically takes ↑ if the initial observation was “up” and ↓ if it was ”down”. All other observations made along the corridor do not matter – they are just noise. InFig 2E, we show the negative log-likelihood of the inferred FSCs with a different numberMof internal states. WithM= 5, the FSC perfectly reconstructs the agent’s behavior (Fig 2f), whereas with fewer internal states the controller fails to memorize and/or make the required associations. The node () represents the “blank slate” internal state of the agent as it enters the maze. Then, the FSC branches to one of two “corridor states” depending on the initial observation, and there it remains until the junction is reached.remembers the first “up” observation, so that the corresponding policy has a slight bias in the up direction. After the “end” observation, the FSC switches fromto, where the up action is taken deterministically. The other states,and, represent instead the first “down” observation, allowing the FSC to perfectly match the agent’s behavior. Finally, we consider a synthetic chemotactic agent that moves in a one-dimensional concentration gradient. At each time, the agent may detect the presence of a chemical () or not () with a probability that depends on its position, and can either decide to run – keep on moving in its current direction,– or tumble – randomly reorient without moving,. The agent chooses the next action based on the last two observations according to the following policy. Assigning valuey= 1 to the detection (•) andy= 0 to no detection (), the agent computes the difference between the last two observations. If the difference is positive, then the agent chooses to run; if it is negative it tumbles; and finally, if it is zero, it chooses at random with equal probability (seeFig 2G). At the beginning of the task, we assume that the agent has not detected a chemical in the last two time-steps. We remark that for this agent the histories of actions and observations are fully entangled: the probability of receiving an observation depends on the position, which in turn depends on the entire past history of actions, each of which depends on previous observations, and so on and so forth. InFig 2H, we show that the negative log-likelihood of the inferred FSCs forM= 4 reaches the value expected for the chosen agent. The corresponding FSC is shown inFig 2G. Inspecting the structure of the transitions between internal states, one realizes that the nodes exactly encode the ordered pair of last observations and the FSC works as a shift-register memory: new observations enter to the right of the register while older observations are pushed to the left and eventually out of it. We note in passing that if the internal computation matrices are low-rank, then the likelihood is invariant under some linear mixing of the internal states (which is not just a trivial permutation), leading to multiple FSCs with exactly the same behavior. This lack of identifiability indeed takes place for the chemotactic agent and is discussed in detail in Section D of theS1 Appendix. We stress that this is not a limitation of FSCs, but a general property of any behavioral model, as it signals a genuine symmetry of the task. Any equivalent FSC provides an equally valid description of the agent’s strategy. We now consider an experimental paradigm where rats are trained to perform evidence accumulation during an auditory discrimination task (data from Ref. [44]). During each trial, rats initially insert their nose into a central point for a fixed amount of time (, signalled by a light-emitting diode). During this time, two trains of randomly timed auditory clicks are played simultaneously, both to the left and to the right of the rat, with different click rates and total durations. At the end of the fixation period, the rat decides to poke either to the left or to the right, and receives a water reward if it turns towards the side where more clicks were played (Fig 3A). The sum of the left and right click rates is fixed at, with subsequent trials varying in difficulty depending on the ratio of the click rates. (a) Sketch of the experiment. During each trial, rats fixate their nose into a central port for. During this time, corresponding to the action “listen”, trains of randomly timed auditory clicks are played both to the left and to the right, with different frequencies and durations (betweenand) which represent three observations: left, right, or silence (“” in the figure). After nose fixation ends, the rat receives an “end” observation, and pokes in one of the side ports (actions ← or →). If it turns to the side where more clicks have been played, it receives a water reward. Data of approximately 4500 trials (half for training, half for validation) from [44]. (b) Negative log-likelihood over the validation set of the inferred FSCs with different numbers of internal states. For, a two-layer structure, which reproduces the structure of the experiments, was imposed a priori to speed up convergence. Inset: negative log-likelihood over the training set during inference. (c) At, FSCs (grayscale dots) accurately reproduce rats’ performance (dashed sand line, averaged across trials and individuals) in choosing right as a function of the difference between the number of clicks. A further improvement can be seen forM= 5 onward. (d) Inferred FSC withM= 2 internal states. The FSC learns to take a ← or → action only after the “end” observation, but cannot perform any computation due to the lack of internal states, and thus must choose at random. (e) Inferred FSC withM= 4 internal states and a fixed two-layer structure. The two internal states on the top (,) are responsible for taking the “listen” action, acting as a computation layer where the agent accumulates evidence. The two at the bottom (,) decide whether to go left or right. The stateremains unchanged if a click is heard on the left or no click is heard, whereas it switches towith a probability. If “end” is heard, the trial ends andswitches with probability one to, where the agent chooses to go to the left. The situation is reversed for the statesand, the latter of which goes to the right. (f) It can be shown that this FSC is computing an exponential smoothing of the difference between right and left observationswith a smoothing factor(Eq. (3). Indeed, the fraction of times that the rats choose to go right is well-predicted by its value at the end of the trial. Sinceis captured by the internal state, the FSC prediction simply corresponds to the bisector of the plane. (a) Sketch of the experiment. During each trial, rats fixate their nose into a central port for. During this time, corresponding to the action “listen”, trains of randomly timed auditory clicks are played both to the left and to the right, with different frequencies and durations (betweenand) which represent three observations: left, right, or silence (“” in the figure). After nose fixation ends, the rat receives an “end” observation, and pokes in one of the side ports (actions ← or →). If it turns to the side where more clicks have been played, it receives a water reward. Data of approximately 4500 trials (half for training, half for validation) from [44]. (b) Negative log-likelihood over the validation set of the inferred FSCs with different numbers of internal states. For, a two-layer structure, which reproduces the structure of the experiments, was imposed a priori to speed up convergence. Inset: negative log-likelihood over the training set during inference. (c) At, FSCs (grayscale dots) accurately reproduce rats’ performance (dashed sand line, averaged across trials and individuals) in choosing right as a function of the difference between the number of clicks. A further improvement can be seen forM= 5 onward. (d) Inferred FSC withM= 2 internal states. The FSC learns to take a ← or → action only after the “end” observation, but cannot perform any computation due to the lack of internal states, and thus must choose at random. (e) Inferred FSC withM= 4 internal states and a fixed two-layer structure. The two internal states on the top (,) are responsible for taking the “listen” action, acting as a computation layer where the agent accumulates evidence. The two at the bottom (,) decide whether to go left or right. The stateremains unchanged if a click is heard on the left or no click is heard, whereas it switches towith a probability. If “end” is heard, the trial ends andswitches with probability one to, where the agent chooses to go to the left. The situation is reversed for the statesand, the latter of which goes to the right. (f) It can be shown that this FSC is computing an exponential smoothing of the difference between right and left observationswith a smoothing factor(Eq. (3). Indeed, the fraction of times that the rats choose to go right is well-predicted by its value at the end of the trial. Sinceis captured by the internal state, the FSC prediction simply corresponds to the bisector of the plane. https://doi.org/10.1371/journal.pcbi.1014585.g003 From the behavioral trajectories, we can discretize time (we set) and define four different observations and three actions. At each timestep, as long as the light is on, the agent may only take the action “listen” and receive a “left” () or “right” (y= 1) observation if it hears a click on the corresponding side, or a “silence” observation (y= 0) if it hears neither. When the light-emitting diode turns off, the agent receives an “end” observation and can take one of the two actions “go left” or “go right”. Among the data in [44], we extracted the behavioral data of three rats, each across two different days. To average out noise in the rats’ behavior and accurately sample different task difficulties, we retained approximately 4500 trials, with 50% used for inferring the FSC parameters and 50% to validate their performance. InFig 3B, we show the negative log-likelihood of the best-performing FSCs for different numbers of internal statesM. Remarkably, we find a sharp decrease atM= 4, after which the performance quickly plateaus. Indeed, the FSCs withclosely reproduce the observed rats’ performance (Fig 3C) with slight deviations visible only forM= 4. This signals that an FSC with a relatively small number of internal states is expressive enough to reproduce the behavioral trajectories. WithM= 2 andM= 3, the inferred FSCs are trivial, as their limited internal structure can only reproduce the fixation period during which rats listen until the “end” cue (seeFig 3dand Section D inS1 Appendix). For example, withM= 2, no computation can be done in the first internal state because “left” and “right” observations always circle back to this state. When “end” is observed, the FSC transitions to the other state and can only choose “go right” or “go left” at random, regardless of the past history. This model can only reproduce the average final decision across all data, and therefore cannot accumulate past evidence. Inferred FSCs withM= 3 nodes similarly fall short. The structure of the interactions for this task hints at a layered organization of the internal states: a computation layer where the agent can only take the “listen” action, and a decision layer responsible for taking the “go” actions and accessible only upon receiving the ”end” signal. For computation to be possible, at least two nodes in the upper layer are required, whereas two are both necessary and sufficient for the final decision. We therefore focused on FSC architectures withcomputation layer nodes and 2 decision layer nodes, imposing this structure a priori for inference efficiency. A posteriori, we then check that these structures are indeed those that minimize the negative log-likelihood. In Section C of theS1 Appendix, we also show that the same FSCs can be recovered using a curriculum learning strategy, which allows the inference procedure to discover the two-layer structure directly from data. Let us first focus on the simplest case,M= 4, shown inFig 3E. We reiterate that, with the layered structure induced by the random interruption, only the two internal states of the top layer are used for internal computation. Despite the small number of states, this FSC is able to achieve a high accuracy in reconstructing behavior, with thestate encoding for left clicks, andfor right ones. The simplicity of the model allows to compute analytically the predicted probability of choosing an action at the end of the trial,, which obeys the recursion relation with. Note thatis an exponential smoothing of the difference between right and left observations, with smoothing factor. Here,is the probability of switching between the two upper memories (see Section D inS1 Appendix). InFig 3F, we show that indeed the fraction of trials in which the rats choose to go right is well predicted by the value ofat the end of the trial. Thus, the simplest nontrivial FSC withM= 4 interprets the decision-making process as the following algorithmic procedure: compute the exponentially recency-weighted difference of right and left clicks and, at the end of the trial, bias the choice proportionally to this quantity. We now seek to understand how the FSC withM= 5 improves upon the previous one, allowing it to reproduce the rats’ behavior with greater accuracy and to saturate the likelihood (seeFigs 3Band3C). InFig 4A, we draw the inferred FSC with five internal states. Similarly to the previous case, we find two nodes in the computation layer (and) that encode the prevalence of left and right clicks, respectively. Once the trial ends, they deterministically transition to the corresponding state in the decision layer (and) where the agent goes to the left and to the right, respectively. However, the transitions betweenandare now mediated by an intermediate node. If the “end” observation is received while the FSC is in this state, it transitions to the decision layer with a slight left bias, reflecting an overall observed preference by the rats. Importantly, thestate is also the one where the FSC is initialized. (a) Similarly to the case ofM= 4, the leftmost () and rightmost () internal states in the computation layer are responsible for encoding the prevalence of left and right clicks, respectively, and go in the corresponding direction once the trial ends (respectivelyand). Importantly, the switch between them is regulated by theinternal state, which transitions to the left or the right when the corresponding click is heard with a probability. The FSC is initialized at this internal state. If the trial ends in thestate, it transitions to eitherorapproximately at random (0.56 and 0.44, respectively). (b) Probability of occupying one of the upper internal states (,,) right before the trial ends, averaged over the observation sequences in the data, as a function of the difference between right and left clicks. The central state is mostly active in an uncertainty region where the difference in the number of clicks is close to zero. (c-d) For a fixed observation sequencey0:t, the probability of occupying a given internal statemay be different between the FSC withM= 5 (colored dots and solid lines) and withM= 4 (dashed lines). In particular, when a set of congruent observations is followed by incongruent ones (e.g., a series of “right” observations first and then a series of “left” ones), the central state acts as a buffer and maintains the occupancy of the right state at higher levels. This reflects in a slower and delayed decay of the probability of taking the action “go right” if the trial were to end at timet(bottom panel). (a) Similarly to the case ofM= 4, the leftmost () and rightmost () internal states in the computation layer are responsible for encoding the prevalence of left and right clicks, respectively, and go in the corresponding direction once the trial ends (respectivelyand). Importantly, the switch between them is regulated by theinternal state, which transitions to the left or the right when the corresponding click is heard with a probability. The FSC is initialized at this internal state. If the trial ends in thestate, it transitions to eitherorapproximately at random (0.56 and 0.44, respectively). (b) Probability of occupying one of the upper internal states (,,) right before the trial ends, averaged over the observation sequences in the data, as a function of the difference between right and left clicks. The central state is mostly active in an uncertainty region where the difference in the number of clicks is close to zero. (c-d) For a fixed observation sequencey0:t, the probability of occupying a given internal statemay be different between the FSC withM= 5 (colored dots and solid lines) and withM= 4 (dashed lines). In particular, when a set of congruent observations is followed by incongruent ones (e.g., a series of “right” observations first and then a series of “left” ones), the central state acts as a buffer and maintains the occupancy of the right state at higher levels. This reflects in a slower and delayed decay of the probability of taking the action “go right” if the trial were to end at timet(bottom panel). https://doi.org/10.1371/journal.pcbi.1014585.g004 InFig 4B, we plot the probability of occupying one of the computation states (,,) right before the trial ends as a function of the difference between the number of clicks heard on the right and on the left. When most observations are from the right, the FSC activates thestate, and vice versa for thestate. Crucially, the occupation of the middlestate peaks when the trial ends in a region of uncertainty, where the difference between clicks is close to zero. This intermediate node plays an important role. InFig 4C, we consider a fixed observation sequencey0:Tfrom the data, characterized first by a series of “right” observations followed by noisy ones where the rat mostly hears “left”. We compute the probability of occupying a given internal state at time,, and compare it with theM= 4 case. We find that, after an initial transient, both FSCs reach a high level of occupancy of, since only right clicks were heard. However, when the incongruent observations start to appear, thestate acts as a buffer and delays the transition to, which is instead quickly occupied whenM= 4. We can compute the predicted probability of choosing an action at the end of the trial for this FSC as well (see Section D inS1 Appendix). Remarkably, the calculations reveal that the rats’ decision is captured by a leaky competing accumulator model (LCA) [46] with lateral inhibition. Namely, the preference for choosing right is proportional to the difference of two accumulators– favoring stateand, respectively – at the final time. These obey the equations and withthe probability of exiting the central state, andthe probability of exiting one of the extremal states (and). Thus, the passive accumulator – the one that contrasts with the current evidence– inhibits the active one. Due to this lateral inhibition, the probability of taking the right action if the trial were stopped at timetdecays more slowly, allowing the FSC withM= 5 to reproduce the rats’ behavior more accurately and helping identify the side on which more clicks were played, as shown inFig 4D. Notice that the fact thatM= 5 provides a better model forthe rats’ decision-making process is also compatible with previous results showing that individual rats exhibit diverse temporal weighting, represented by theandparameters in the inferred FSC. We now consider a different experiment, where mice are trained to perform a reversal task and make decisions in a changing and uncertain environment (data from [45]). At each step, a mouse can choose from two side ports. One of the two ports has a high probability, e.g., 0.8, of delivering a water reward (the high port), whereas the other has a probability 0.2 of rewarding the mice (the low port). After the completion of each trial, the high and the low ports have a probability 0.02 of switching. This results in blocks of consecutive trials, of variable length, where the environment is predictable, even if uncertain. When the environment switches, however, the mouse has to swiftly adapt its strategy to the change (Fig 5A). (a) Sketch of the experiment. Mice are trained to select one of two side ports, representing two possible actions. Each port has a different probability of giving a water reward, either 80% (the high port) or 20% (the low port), corresponding to the two possible observations. At each time, the high port may randomly switch with a 2% probability, so that mice have to adjust their choices accordingly. Data of 159 behavioral trajectories (126 for training, and 33 for validation) from [45]. (b) Negative log-likelihood over the validation set of the inferred FSCs with different numbers of internal states. Pentagons represent the FSCs with minimum loss across restarts of the inference procedure, and selected restarts are represented by dots. Inset: evolution of the negative log-likelihood over the training set during inference. (c-d) The FSCs reproduce both the probability of selecting the higher reward port () and the probability of switching port selection when the high port changes (), with an accuracy that increases with the number of internal statesMand saturates fromM= 4. (e) Inferred FSC withM= 2,M= 3, andM= 4. From an internal state, green arrows represent the probability of taking the action left, and red arrows the probability of choosing right. Transitions between the states depend on the observation, either when a reward is received (blue) or not (black). For ease of visualization, actions with low probability and the corresponding state transitions have a higher transparency. In all cases, we can identify two extremal states (and) primarily encoding the left or right action, respectively. ForM> 2, the central states (and) buffer the transitions between the extremal ones, similarly toFig 4. (f) Marginalizing over the actions emphasizes the linear structure of the FSC. ForM> 2, the FSCs have a high probability of transitioning to the central states in the absence of a reward, whereas they drift towards the extremal states when a reward is obtained. (g-i) Probability of occupying the internal states as a function of time and for a specific sequence of high ports (right, left, right), averaged over 104trajectories. WithM= 3 andM= 4, the central states quickly activate when the port changes, playing the role of switch detectors. This allows the FSC to mimic the mice’s strategy and switch to the high-reward action. (a) Sketch of the experiment. Mice are trained to select one of two side ports, representing two possible actions. Each port has a different probability of giving a water reward, either 80% (the high port) or 20% (the low port), corresponding to the two possible observations. At each time, the high port may randomly switch with a 2% probability, so that mice have to adjust their choices accordingly. Data of 159 behavioral trajectories (126 for training, and 33 for validation) from [45]. (b) Negative log-likelihood over the validation set of the inferred FSCs with different numbers of internal states. Pentagons represent the FSCs with minimum loss across restarts of the inference procedure, and selected restarts are represented by dots. Inset: evolution of the negative log-likelihood over the training set during inference. (c-d) The FSCs reproduce both the probability of selecting the higher reward port () and the probability of switching port selection when the high port changes (), with an accuracy that increases with the number of internal statesMand saturates fromM= 4. (e) Inferred FSC withM= 2,M= 3, andM= 4. From an internal state, green arrows represent the probability of taking the action left, and red arrows the probability of choosing right. Transitions between the states depend on the observation, either when a reward is received (blue) or not (black). For ease of visualization, actions with low probability and the corresponding state transitions have a higher transparency. In all cases, we can identify two extremal states (and) primarily encoding the left or right action, respectively. ForM> 2, the central states (and) buffer the transitions between the extremal ones, similarly toFig 4. (f) Marginalizing over the actions emphasizes the linear structure of the FSC. ForM> 2, the FSCs have a high probability of transitioning to the central states in the absence of a reward, whereas they drift towards the extremal states when a reward is obtained. (g-i) Probability of occupying the internal states as a function of time and for a specific sequence of high ports (right, left, right), averaged over 104trajectories. WithM= 3 andM= 4, the central states quickly activate when the port changes, playing the role of switch detectors. This allows the FSC to mimic the mice’s strategy and switch to the high-reward action. https://doi.org/10.1371/journal.pcbi.1014585.g005 From the behavioral trajectories, we can immediately identify two possible actions – “choose left” or “choose right” – and two observations – “water” and “no water”. Importantly, and differently from the previous case, the observations received by the agent are now explicitly dependent on the action it has taken. Furthermore, since any observation can follow from an action, we do not impose any a priori structure on the FSC. We consider the 159 trials reported in [45], with 80% of them used for inference and 20% for validation. InFig 5B, we plot the negative log-likelihood of the inferred FSC for different numbers of internal states. Once more, we find that the likelihood plateaus already at relatively small values ofM. However, at variance with the previous section, we find pronounced improvements when moving fromM= 2 toM= 4. In Section B of theS1 Appendix, we explicitly compare the inferred FSCs with different types of hidden Markov models. We find that, while some HMMs can infer the immediate reward-to-action mapping, they cannot reproduce the mice’s commitment due to their inability to encode past actions in their internal states. To quantitatively assess the ability of the FSC in capturing the decision process of the mice, we compute the probability of selecting the high reward portand the probability of switching port selection when the high port changes, both as a function of the time from a switch [45]. We find that the inferred FSCs withM= 4 or above are able to reproduce the data very accurately. Remarkably, even forM= 2 andM= 3, the FSCs are still able to capture the timescales at which mice switch their behavior on average. InFig 5E, we draw the structure of the inferred FSCs withM= 2, 3, 4. In all cases, we find two extremal states (and) where the probability of choosing left and right, respectively, is significantly higher. This suggests that these states encode the agent’s confidence that the corresponding side is the one with a higher chance of delivering the reward. ForM= 3, a central stateemerges, from which the action is taken almost at random. Similarly to the previous example, this state buffers the transition between the extremal ones, delaying the commitment to a particular action. ForM= 4, this effect is ascribed to two separate central states (,) where the policy still allows taking both actions, but is skewed towards the left and right ones, respectively. By marginalizing over the actions, the transitions between the internal states at fixed observation emphasize that all FSCs have a linear structure (Fig 5F). That is, the inferred internal computationgdrives the agent towards the extremal states when a reward is obtained, whereas it transitions to the central ones in the absence of a reward. This linear structure is maintained even for largerM, suggestingthat it can consistently model the internal decision process of the mice. The diminishing return in terms of increased accuracy fromM= 4 onward is further highlighted by the fact that at largeMsome inferred internal states become disconnected from the rest, signaling that they are not needed to reproduce the observed behavior (see Section C inS1 Appendix). We now focus on the mechanisms that enable FSCs with largerMto better represent mice behavior. To this end, we generate behavioral trajectories from the FSCs for a fixed sequence of high ports (right, left, right) and track the probability of occupying a given internal state (Figs 5G,5Hand5I. ForM= 2, the FSC is able to correctly follow the sequence of environments, but the probability of occupying the state corresponding to the worst action remains high (e.g.,when the high port is the right one). This results in a lower probability of taking the best action, as also shown inFig 5C. ForM= 3, accuracy substantially increases thanks to the middle state. InFig 5Hwe show that the probabilityp(mt=) peaks right after the high port changes. Hence,acts as a switch detector, quickly activating when the number of rewards decreases more than expected in the previous state. The FSC withM= 4 is able to further improve its model of behavior thanks to the emergence of two dedicated switch detectors. As we show inFig 5I, theactivates when the high port changes from the right to the left, whereasis responsible for detecting the left-to-right switch. The policiesπ(a|) andπ(a|) are still dominated by the right and left action, respectively, but are stochastic, allowing these states to probe the opposite action. The presence of these two switch detectors allows the accurate reconstruction of the mice behavior. The inference of decision-making processes hinges upon the availability of data about the interaction between agent and environment in the form of histories of actions and observations. In the cases discussed above, the identification of actions and observations is straightforward, but more complex situations command more attention. Several technical tools are now available to preprocess raw behavioral data and isolate suitable candidates for actions and observations [47–50]. In general, the key step resides in the choice of the level of description that the agent model wants to address. For instance, in a human navigation task, there is a coexistence of low-level actions and observations – like taking a step in some direction and noticing the presence of obstacles – and high-level actions and observations – like moving from one room to another and reading direction signs. Further, here we focused on the setting where observationsare received by the agent after actions, representing agents that actively interact with the environment. In this case, actions can serve a dual purpose: achieving a goal – such as moving towards a target or obtaining food – and gathering information – for instance, revealing occluded objects or exploring the environment. Several, if not most, sensory processes can be described at a scale where they belong to this category [51,52]. Nevertheless, other choices are possible (see Section B inS1 Appendix), and our framework can be readily applied to them as well. We stress that we did not make any attempt at modeling the environment, nor did we require any prior information about it. However, in the event that some knowledge about the nature of the task is available beforehand, it can be used to appropriately design the agent model. In all these cases, one can always check a posteriori that the negative log-likelihood of the constrained FSCs is lower than that inferred without any structural assumption. For example, in the evidence accumulation task above, the presence of an external signal that ends the acquisition of information naturally suggests adopting a two-layer topology for the FSC. In the case of a free-response experiment in which the agent freely decides when to go right or left (see, e.g., [53]), a simple linear topology may turn out to be sufficient. When an environment model is known, such as in the trial-based experimental settings considered here, FSCs can be used as generative models when a model of the environment is known. In some cases – such as the evidence accumulation tasks for rats – observations are independent of actions, and thus the inferred FSCs can generate actions in response to arbitrarily chosen sequences of observations. In general, however, the generation of behavioral trajectories requires knowledge about how observations depend on actions. In any event, using the inferred FSC in generative mode gives the possibility of predicting the agent’s response to a crafted environment and suggesting ways to steer the agent’s behavior in a desired direction. While our approach is well-suited for those cases where elementary actions and observations are inherently discrete, real tasks are often described by continuous, possibly high-dimensional behavioral variables. In some cases, it may still be possible to segment behavioral data and obtain once more trajectories of discrete behavioral modes – actions – and observations. Different methods may allow one to analyze behavior at different scales, and an interesting question would then be whether the inferred FSCs remain consistent or whether they reflect the specific behavioral scale captured by the segmentation. In the latter case, comparing FSCs may be a helpful tool for understanding the hierarchical structure of behavior: internal states that persist across segmentations would correspond to robust computational motifs, while states that appear only at finer scales would reveal transient or low-level strategies that are averaged out at coarser descriptions. In practice, however, inference becomes increasingly complex as the number of actions and observations increases, which entails an increase in the number of parameters, effectively bounding the number of internal states that can be resolved from finite behavioral datasets. To deal with these scenarios, a possible extension would be to let the agent explicitly filter irrelevant inputs according to its internal state. This would allow the FSC to extract task-relevant observations from raw sensory inputs, reducing the effective number of observations that each internal state must account for. A complementary approach would be to take into account the continuous nature of actions and observations by resorting to function approximation in order to represent the agent with a manageable number of parameters. For instance, one could replace the parametrization in the Methods with a more general softmax with appropriately selected feature vectors, or endow each internal state with a learnable dynamical model, similarly to other state space models [54]. While a discussion of these approaches goes beyond the scope of this paper, we can report some success in some selected tasks. Further work still remains to be done in this direction. In general, we stress that the minimalist design of agent models like FSCs is especially useful if a high-level description of behavior is sought. Indeed, an agent model based on low-level actions and observations alone typically requires a larger number of internal states to account for all the corresponding low-level computations, whereas fewer internal states often suffice at the higher level of description. Sometimes, however, it is impossible to avoid the granularity of the lowest levels of description, and this may lead to an unsustainable growth of complexity of the agent model. In these cases, a good compromise is achieved by a hierarchical organization in which different levels are stacked one upon another [55]. For instance, behavior trees [56] are a specific example of hierarchical decision structures widely used in robotics and in game development that could provide inspiration for building parsimonious agent models. In such scenarios, one could imagine inferring directly hierarchical FSCs or behavioral trees to take the granularity of behavior directly into account. Furthermore, it will also be interesting to quantitatively understand in which cases few discrete states can provide good descriptions of behavioral strategies, and when instead other approaches are needed – e.g., behavior can only be described with genuinely continuous internal states. Finally, we can connect FSC and neural computations by their formal correspondence with linear fully gated recurrent units. For the evidence accumulation tasks, this is even more explicit by the finding that the inferred FSC withM= 5 nodes can be interpreted as a leaky competing accumulator model. More generally, the internal states of FSC could be viewed as attractors of a recurrent neural network dynamics. This interpretation is consistent with the fact that recurrent neural networks trained on some task display a fast dynamics of convergence to attractors and slow transitions between them, which can be triggered by external inputs or occur spontaneously. This slow dynamics can be effectively represented by discrete states and transitions [24,28,57–59], and it has been suggested that it underlies flexible decisions and computations through dynamics [27,60]. In this context, linear arrangements of internal states, like the ones that emerge in the analysis of rats’ and mice’s behavior, might then be interpreted as proxies for line attractors [32,61]. The fact that FSC inference appears to be robust even with small numbers of trajectories also opens the avenue to fit agent models directly to single animals and quantify the variability across individuals. In future works, one could pair such individual-based inference with single-trial neural recordings from the same individuals to test whether the inferred internal states correspond to identifiable neural signatures. In this work, we discussed applications to experimental data of rodent behavior, but our approach does not make assumptions about the nature of the agent. Any underlying computational process, be it neural, biochemical, or artificial, as long as it supports a coarse-grained description in terms of discrete states, could be a suitable candidate to be described by an FSC. It would be interesting to explore the applicability of our method to a broader range of biological agents and tasks, including human decision-making. A finite state controller withMinternal states is defined by three probability distributions – the policy, the internal computation, and the initial state distribution – which we parametrize using softmax functions. The policymodels the probability of taking an actionfrom a specific internal state, withand. Hence, whereare the parameters we seek to infer. The internal computationmodels the probability of transitioning to a statefrom a statem, conditioned on the actionataken by the agent and the observationit received, with. As before, where. Finally, the initial internal state of the FSC is specified by with. Therefore, the total number of parameters isM(1 +A+MAY). Finite state controllers may be viewed as a generalized type of state space models (SSMs) [62] with recurrent transitions. One of the crucial features of FSCs is that transitions between internal states explicitly depend on both actions and observations. This recurrence allows them to encode previous actions and observations in their internal states. InFig 6we explicitly compare FSCs (Fig 6A) with different types of hidden Markov models (HMMs). (a) The FSC model of the agent, where transitions between internal states are fully recurrent - i.e., they depend on the previous observation, action, and state. (b) In a hidden Markov model (HMM), observations are ignored, and transitions between internal states are autonomous. The policy is known as the emission probability, and solely depends on the current internal state (as in FSCs). (c) In a generalized linear model HMM (GLM-HMM), the transitions between internal states remain autonomous, but the policy depends explicitly on the observation via a generalized linear model [22]. (d) In a general input-output HMM [20], both internal transitions and the policy depend on the observation, but not on the emissions, as in FSCs. (a) The FSC model of the agent, where transitions between internal states are fully recurrent - i.e., they depend on the previous observation, action, and state. (b) In a hidden Markov model (HMM), observations are ignored, and transitions between internal states are autonomous. The policy is known as the emission probability, and solely depends on the current internal state (as in FSCs). (c) In a generalized linear model HMM (GLM-HMM), the transitions between internal states remain autonomous, but the policy depends explicitly on the observation via a generalized linear model [22]. (d) In a general input-output HMM [20], both internal transitions and the policy depend on the observation, but not on the emissions, as in FSCs. https://doi.org/10.1371/journal.pcbi.1014585.g006 In a standard HMM (Fig 6B) transitions between internal states are autonomous and depend only on the previous state, while the policy – also known as the emission probability – depends on the current internal state, as in FSCs. Thus, observations do not influence transitions. A possible generalization is HMMs where emissions are parametrized by a generalized linear model (GLM-HMMs), so that the policy explicitly depends on observations [22] (Fig 6C). More generally, in input-output hidden Markov models (IO-HMMs), the dependence on observations enters both the emission probability and the internal computation [20] (Fig 6D). In Section B of theS1 Appendix, we show that these types of models – as well as extended versions of GLM-HMMs whose policy depends on the previous actions as well – fail to capture the mice behavior in the reversal task due to the fact that the dynamics of the internal states is disentangled from that of the emissions, which is instead a key dependency in FSCs. The goal is to identify the FSC which is as “close” as possible to the true unknown agent, i.e., solve the minimization problem where FSC is a shorthand for the set of parameters. Above,is the expectation over the distribution of trajectories of actions and observationsgenerated by the interaction of the agent and the environment,Dis the Kullback-Leibler divergence, and where the policy, the internal computation, and the initial distribution are parametrized as described in the previous section. Noting that and replacing the expectation value with the empirical distribution over a set ofNbehavioral trajectoriesfor, one arrives at the maximum-likelihood problem (1) The main idea of MAPSO is sketched inFig 7. First,FSCs are initialized with random parameters. We can think of each FSC as a particlein thespace. The parameters of each particle are updated according to a swarming dynamics, where the velocity of each particle is controlled by three different terms. First, an inertia term decreases the velocity with the number of epochs. The second term is a so-called cognitive coefficient, where the velocity vector of thep-th particle evolves to point towards the best set of parameters ever seen by the particle (). Finally, the last term is a social coefficient, where the velocity tends to point towards the best set of parameters ever seen by one of the neighbors of the particles (), where the neighbors are specified by a network of interaction that changes as the algorithm progresses. At the beginning, during a “global phase” all particles interact with one another. Then, a “local metric phase” takes place, where at each step of the algorithm, each particle interacts only with a finite number of neighbors defined by the Euclidean distance between them. The number of neighbors decreases until the particles interact only with their nearest neighbors, before increasing back. The algorithm ends with a global phase to guarantee that all particles share the same set of best parameters. A given number of FSCs are initialized randomly, representing a swarm of particles in the high-dimensional parameter space of the policy, computation, and initial distribution. During an initial global phase, their parameters are updated according to an all-to-all swarming dynamics, where each FSC tends to move towards the best set of parameters it has seen so far, as well as towards the best set of parameters ever found by the swarm. After each step of the swarm evolution, this global best is mutated along a random direction and is kept if the mutation improves the likelihood, and the swarm dynamics is updated according to pre-defined adaptive rules [42]. The global phase is followed by a local metric phase, where the swarm dynamics only involve a finite number of neighbors defined by the Euclidean distance between the particles. During this phase, the FSCs evolve towards the local best set of parameters, i.e., the best set of parameters ever found by the neighbors. The number of neighbors decreases before increasing back, and the algorithm ends with a second global phase. A given number of FSCs are initialized randomly, representing a swarm of particles in the high-dimensional parameter space of the policy, computation, and initial distribution. During an initial global phase, their parameters are updated according to an all-to-all swarming dynamics, where each FSC tends to move towards the best set of parameters it has seen so far, as well as towards the best set of parameters ever found by the swarm. After each step of the swarm evolution, this global best is mutated along a random direction and is kept if the mutation improves the likelihood, and the swarm dynamics is updated according to pre-defined adaptive rules [42]. The global phase is followed by a local metric phase, where the swarm dynamics only involve a finite number of neighbors defined by the Euclidean distance between the particles. During this phase, the FSCs evolve towards the local best set of parameters, i.e., the best set of parameters ever found by the neighbors. The number of neighbors decreases before increasing back, and the algorithm ends with a second global phase. https://doi.org/10.1371/journal.pcbi.1014585.g007 All the parameters governing the relative strength of these terms are adaptively set according to the fuzzy rules introduced in [42]. Furthermore, at each step, the swarming dynamics is hybridized with a genetic algorithm, where eachis mutated along a random direction and retained if it reduces the corresponding negative log-likelihood. MAPSO runs for a fixed number of epochs, where each epoch corresponds to a velocity update, a particle update, a parameter strength update, a computation of the metric distances between the particles and the resulting topology, and a mutation step. See Section F of theS1 Appendixfor more details and an analysis of MAPSO robustness against hyperparameter choices. To select the best inferred FSC, we consider a cross-validation scheme. We randomly split the data into a training set (50% of the available trajectories in the rats’ experiments, and 80% of the available trajectories in the mice’s experiments) and a validation set (the remaining trajectories). Due to the random split, we note that this implies that the training and validation sets will have trials that, in general, are interdispersed in time. In scenarios where long-timescale effects such as satiation may be important, more careful splits may be chosen. After running MAPSO for a fixed number of epochs over the training set, the best set of parameters ever found by a particle is returned. These parameters correspond todue to the final global phase where all particles interact with one another. Since the optimization problem is non-convex, we run the training procedure multiple times for a given number of internal statesM, each with a random set of initial conditions for MAPSO. Then, we compute the negative log-likelihood over the validation set of each of the inferred FSCs, and retain the one with the smallest. In this work, the initial parameters were extracted from a multivariate Gaussian distribution centered around zero and with diagonal covariance, but other choices are possible to bias the initial search in specific regions of the parameter space. We also note that MAPSO allows for other schedules, such as sequential restarts, where the particles’ initial position is not independent but is centered around the best parameters previously found with varying variance. See Section F of theS1 Appendixfor an analysis of MAPSO’s results across restarts. Additional descriptions of finite state controllers, the inference procedure, comparison with other models and methods, theoretical derivations, and figures. https://doi.org/10.1371/journal.pcbi.1014585.s001 (PDF) We acknowledge useful discussions with E. Panizon and K. V. B.F Verano in the early stages of this work.
→ Apri originale
El 52% de los gestores de flotas admite que la incertidumbre regulatoria condiciona sus inversiones - www.diariodelpuerto.com
📰 www.diariodelpuerto.com Media 📅 2026-08-07 es
El 52% de los gestores de flotas admite que la incertidumbre regulatoria condiciona sus inversiones www.diariodelpuerto.com
→ Apri originale
Wall Street apre positiva, Dj a +0,14% e Nasdaq a + 0,77% - La Gazzetta del Mezzogiorno
📰 La Gazzetta del Mezzogiorno Media 📅 2026-08-07 it
Wall Street apre positiva, Dj a +0,14% e Nasdaq a + 0,77% La Gazzetta del Mezzogiorno
→ Apri originale
Un alcalde y otras dos personas detenidas por el incendio cerca de Atenas - RFI
📰 RFI 📅 2026-08-07 es
Un alcalde y otras dos personas detenidas por el incendio cerca de Atenas RFI
Atenas (AFP) –La justicia griega puso en detención preventiva a tres personas, entre ellas el alcalde de una localidad al oeste de Atenas, en relación con el incendio forestal que arrasó en esa región más de 11.000 hectáreas la pasada semana. Primera modificación:07/08/2026 - 15:44 Los investigadores creen que el incendio se inició el 31 de julio a causa de una chispa procedente de los cables eléctricos de un parque eólico situado cerca de Agios Vasilios, un pueblo costero a 85 kilómetros al noroeste de Atenas. El alcalde de la localidad de Stylida (a 140 km al noroeste de la capital), Giannis Apostolou, es propietario de una empresa de instalaciones eléctricas vinculada al parque eólico. Los investigadores le acusan, así como a un contratista de su empresa y al propio propietario del parque eólico, de "incendio intencionado en un bosque" y de "daños a instalaciones de utilidad pública", según medios locales. Se enfrentan a penas de más de diez años de cárcel. Según la agencia de noticias griega ANA, las autoridades han decidido suspender las operaciones del parque eólico, situado en la frontera entre las regiones de Beocia y Ática. Los fuertes vientos, con rachas que alcanzaron los 130 km/h el pasado fin de semana, propagaron el fuego por el monte Citerón, cerca de Atenas, y quemaron más de 11.000 hectáreas en cinco días, provocando un "grave daño medioambiental", lamentó el miércoles el primer ministro griego, Kyriakos Mitsotakis. "Haremos todo lo que esté en nuestra mano para favorecer la regeneración natural", aseguró. El propietario de las turbinas eólicas está acusado de no haber limpiado la vegetación. Según ANA, el 27 de junio ya se había producido un incendio en la misma línea eléctrica. Cuando se instaló el parque eólico, hace más de 10 años, los vecinos intentaron sin éxito impedirlo mediante acciones legales. Además del bosque, los matorrales y los olivares afectados por el incendio, en la localidad costera de Porto Germeno, a orillas del golfo de Corinto, resultaron dañadas más de 120 construcciones y viviendas. © 2026 AFP
→ Apri originale
Guasto sulla rete idrica, disagi per molte utenze. Hera: "Intervento risolto. Presto il ripristino dell'acqua" - RavennaToday
📰 RavennaToday Media 📅 2026-08-07 it
Guasto sulla rete idrica, disagi per molte utenze. Hera: "Intervento risolto. Presto il ripristino dell'acqua" RavennaToday
Un intervento urgente sulla condotta idrica principale che alimenta la frazione di Classe ha creato disagi nella gioranta di oggi, venerdì, a molti cittadini. Hera informa che i lavori di riparazione sulla rete idrica sono stati completati e l'erogazione dell'acqua è in fase di ripristino su tutto il territorio interessato. Nel pomeriggio i tecnici sono stati al lavoro in via Romea Sud, all'angolo con via Pescara, a causa di una rottura consistente e potenzialmente pericolosa per la circolazione del traffico pesante. Il guasto ha riguardato una tubazione importante posizionata in prossimità di un nodo di chiusura. Per eseguire le operazioni di ripristino in sicurezza, si era reso necessario sospendere la fornitura idrica a gran parte della frazione, con un impatto stimato su almeno 300 utenze. Le squadre operative erano sul posto per contenere al minimo i disagi e garantire che il servizio venga ripristinato in tempi brevi. L'azienda assicura il massimo impegno per ridurre al minimo la durata dell'interruzione e ricorda che, per segnalazioni guasti, rotture o emergenze sui servizi acqua, fognature e depurazione, è attivo 24 ore su 24, sette giorni su sette, il numero di pronto intervento gratuito 800.713.900. ScegliRavennaTodaycome fonte preferita su Google News Leggi tutte leinchieste e gli approfondimenti Vuoi navigaresenza pubblicità? Scopri come fare © Riproduzione riservata
→ Apri originale
🏠