{"id":"https://openalex.org/W7138913316","doi":"https://doi.org/10.1609/aaai.v40i41.40794","title":"When Instinct Guides and Insight Grounds: Staged RL Training for LLM Agents","display_name":"When Instinct Guides and Insight Grounds: Staged RL Training for LLM Agents","publication_year":2026,"publication_date":"2026-03-14","ids":{"openalex":"https://openalex.org/W7138913316","doi":"https://doi.org/10.1609/aaai.v40i41.40794"},"language":"en","primary_location":{"id":"doi:10.1609/aaai.v40i41.40794","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i41.40794","pdf_url":"https://ojs.aaai.org/index.php/AAAI/article/download/40794/44755","source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"diamond","oa_url":"https://ojs.aaai.org/index.php/AAAI/article/download/40794/44755","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100759406","display_name":"Zeyu Zhang","orcid":"https://orcid.org/0009-0007-2124-7802"},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zijing Zhang","raw_affiliation_strings":["Peking University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Peking University","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5061137603","display_name":"Boning Zhang","orcid":null},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Boning Zhang","raw_affiliation_strings":["Institute of Automation, Chinese Academy of Sciences"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Automation, Chinese Academy of Sciences","institution_ids":["https://openalex.org/I19820366"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":"40","issue":"41","first_page":"34906","last_page":"34914"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.23280000686645508,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.23280000686645508,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.19509999454021454,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.13359999656677246,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.5942999720573425},{"id":"https://openalex.org/keywords/robustness","display_name":"Robustness (evolution)","score":0.49300000071525574},{"id":"https://openalex.org/keywords/action","display_name":"Action (physics)","score":0.4715999960899353},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.46209999918937683},{"id":"https://openalex.org/keywords/instinct","display_name":"Instinct","score":0.4390999972820282},{"id":"https://openalex.org/keywords/phenomenon","display_name":"Phenomenon","score":0.3919000029563904}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6211000084877014},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.5942999720573425},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.49300000071525574},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.4715999960899353},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4675999879837036},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.46209999918937683},{"id":"https://openalex.org/C3232509","wikidata":"https://www.wikidata.org/wiki/Q18237","display_name":"Instinct","level":2,"score":0.4390999972820282},{"id":"https://openalex.org/C50335755","wikidata":"https://www.wikidata.org/wiki/Q483247","display_name":"Phenomenon","level":2,"score":0.3919000029563904},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.3499999940395355},{"id":"https://openalex.org/C106301342","wikidata":"https://www.wikidata.org/wiki/Q4117933","display_name":"Entropy (arrow of time)","level":2,"score":0.3156000077724457},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.289000004529953},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.28630000352859497},{"id":"https://openalex.org/C13687954","wikidata":"https://www.wikidata.org/wiki/Q4826847","display_name":"Autonomous agent","level":2,"score":0.28060001134872437},{"id":"https://openalex.org/C182306322","wikidata":"https://www.wikidata.org/wiki/Q1779371","display_name":"Order (exchange)","level":2,"score":0.2782000005245209},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.27090001106262207}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1609/aaai.v40i41.40794","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i41.40794","pdf_url":"https://ojs.aaai.org/index.php/AAAI/article/download/40794/44755","source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},{"id":"pmh:oai:ojs.aaai.org:article/40794","is_oa":false,"landing_page_url":"https://ojs.aaai.org/index.php/AAAI/article/view/40794","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"2159-5399","raw_type":"info:eu-repo/semantics/article"}],"best_oa_location":{"id":"doi:10.1609/aaai.v40i41.40794","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i41.40794","pdf_url":"https://ojs.aaai.org/index.php/AAAI/article/download/40794/44755","source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/16","score":0.760253369808197,"display_name":"Peace, Justice and strong institutions"}],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7138913316.pdf","grobid_xml":"https://content.openalex.org/works/W7138913316.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"Language":[1],"Model":[2],"(LLM)":[3],"agents":[4,18],"have":[5],"demonstrated":[6],"strong":[7,174],"potential":[8],"in":[9,28,43,80,185],"complex,":[10],"interactive":[11],"decision-making":[12],"tasks.":[13],"However,":[14],"when":[15],"training":[16,101,140,170,192],"LLM":[17,37,190],"end-to-end":[19],"with":[20],"reinforcement":[21],"learning":[22],"(RL),":[23],"efficiently":[24],"optimizing":[25],"agent":[26,38,115,191],"policies":[27],"dynamic":[29],"environments":[30],"remains":[31],"a":[32,44,57,68,98,195],"significant":[33],"challenge.":[34],"Existing":[35],"RL-based":[36],"paradigms":[39],"commonly":[40],"organize":[41],"interactions":[42],"cycle":[45],"where":[46,63],"reasoning":[47,69,204],"is":[48],"followed":[49],"by":[50,76,126],"action.":[51],"In":[52,103,129],"our":[53],"work,":[54],"we":[55,59,94,107,133],"observe":[56],"phenomenon":[58],"call":[60],"Exploration":[61],"Contraction,":[62],"the":[64,72,104,109,114,130,135,150,180,186,200],"explicit":[65,203],"introduction":[66],"of":[67,74,182,202],"stage":[70],"reduces":[71],"diversity":[73],"actions\u2014quantified":[75],"lower":[77],"action":[78,183],"entropy\u2014which":[79],"turn":[81],"limits":[82],"exploration":[83,124,160],"and":[84,141,145,152,168,193],"leads":[85],"to":[86,116,120,173,198],"premature":[87],"policy":[88],"convergence.":[89],"To":[90],"address":[91],"this":[92],"limitation,":[93],"propose":[95],"Act-before-Reasoning":[96],"(ActRe),":[97],"two-stage":[99],"RL":[100,175],"framework.":[102],"first":[105],"stage,":[106,132],"reverse":[108],"typical":[110],"rollout":[111],"order,":[112],"prompting":[113],"generate":[117],"actions":[118],"prior":[119],"reasoning,":[121],"which":[122],"encourages":[123],"driven":[125],"model":[127],"intuition.":[128],"second":[131],"restore":[134],"standard":[136],"reasoning-then-action":[137],"order":[138],"for":[139],"evaluation,":[142],"ensuring":[143],"robust":[144],"interpretable":[146],"decision-making.":[147],"Experiments":[148],"on":[149],"ALFWorld":[151],"WebShop":[153],"benchmarks":[154],"show":[155],"that":[156],"ActRe":[157],"effectively":[158],"mitigates":[159],"contraction,":[161],"yielding":[162],"consistently":[163],"higher":[164],"task":[165],"success":[166],"rates":[167],"improved":[169],"robustness":[171],"compared":[172],"baselines.":[176],"Our":[177],"analysis":[178],"underscores":[179],"importance":[181],"entropy":[184],"exploration-exploitation":[187],"trade-off":[188],"during":[189],"provides":[194],"practical":[196],"approach":[197],"maintain":[199],"benefits":[201],"while":[205],"promoting":[206],"sufficient":[207],"exploration.":[208]},"counts_by_year":[],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2026-03-20T00:00:00"}