{"id":"https://openalex.org/W7214358361","doi":"https://doi.org/10.48550/arxiv.2609.28963","title":"Back to the Definition: Estimating Step-Level Advantages via Trajectory Graphs for Agentic Reinforcement Learning","display_name":"Back to the Definition: Estimating Step-Level Advantages via Trajectory Graphs for Agentic Reinforcement Learning","publication_year":2026,"publication_date":"2026-09-24","ids":{"openalex":"https://openalex.org/W7214358361","doi":"https://doi.org/10.48550/arxiv.2609.28963"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.28963","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.28963","pdf_url":"https://arxiv.org/pdf/2609.28963","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2609.28963","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5124148869","display_name":"Xincheng Yao","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yao, Xincheng","raw_affiliation_strings":["School of Information Science and Electronic Engineering , Shanghai Jiao Tong University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Information Science and Electronic Engineering , Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5152814827","display_name":"Haobo Fu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I2250653659","display_name":"Tencent (China)","ror":"https://ror.org/00hhjss72","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250653659"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Fu, Haobo","raw_affiliation_strings":["Tencent AI Platform Department "],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Tencent AI Platform Department ","institution_ids":["https://openalex.org/I2250653659"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5007858277","display_name":"Weiming Liu","orcid":"https://orcid.org/0000-0002-4115-7667","observed_orcids":["https://orcid.org/0000-0002-4115-7667","https://orcid.org/0000-0002-4262-639X"]},"institutions":[{"id":"https://openalex.org/I2250653659","display_name":"Tencent (China)","ror":"https://ror.org/00hhjss72","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250653659"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Liu, Weiming","raw_affiliation_strings":["Tencent AI Platform Department "],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Tencent AI Platform Department ","institution_ids":["https://openalex.org/I2250653659"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5152837760","display_name":"Chongyang Zhang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhang, Chongyang","raw_affiliation_strings":["MoE Key Lab of Artificial Intelligence , AI Institute , Shanghai Jiao Tong University","School of Information Science and Electronic Engineering , Shanghai Jiao Tong University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"MoE Key Lab of Artificial Intelligence , AI Institute , Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]},{"raw_affiliation_string":"School of Information Science and Electronic Engineering , Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]}],"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":1,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.9940999746322632,"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"}},"topics":[{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.9940999746322632,"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":0.0006000000284984708,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.0005000000237487257,"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/agentic-reinforcement-learning","display_name":"agentic reinforcement learning","score":0.7490000128746033},{"id":"https://openalex.org/keywords/group-relative-policy-optimization","display_name":"Group Relative Policy Optimization","score":0.949999988079071},{"id":"https://openalex.org/keywords/credit-assignment","display_name":"credit assignment","score":0.7620000243186951},{"id":"https://openalex.org/keywords/multi-turn-reasoning","display_name":"multi-turn reasoning","score":0.5770000219345093},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.671999990940094}],"concepts":[{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.7613999843597412},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7161999940872192},{"id":"https://openalex.org/C13662910","wikidata":"https://www.wikidata.org/wiki/Q193139","display_name":"Trajectory","level":2,"score":0.5203999876976013},{"id":"https://openalex.org/C132525143","wikidata":"https://www.wikidata.org/wiki/Q141488","display_name":"Graph","level":2,"score":0.5127999782562256},{"id":"https://openalex.org/C62611344","wikidata":"https://www.wikidata.org/wiki/Q1062658","display_name":"Node (physics)","level":2,"score":0.4771000146865845},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.41999998688697815},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.41280001401901245},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.40799999237060547},{"id":"https://openalex.org/C43214815","wikidata":"https://www.wikidata.org/wiki/Q7310987","display_name":"Reliability (semiconductor)","level":3,"score":0.39649999141693115},{"id":"https://openalex.org/C204323151","wikidata":"https://www.wikidata.org/wiki/Q905424","display_name":"Range (aeronautics)","level":2,"score":0.3682999908924103},{"id":"https://openalex.org/C2779913896","wikidata":"https://www.wikidata.org/wiki/Q7063001","display_name":"Notice","level":2,"score":0.36010000109672546},{"id":"https://openalex.org/C148220186","wikidata":"https://www.wikidata.org/wiki/Q7111912","display_name":"Outcome (game theory)","level":2,"score":0.35440000891685486},{"id":"https://openalex.org/C14646407","wikidata":"https://www.wikidata.org/wiki/Q1430750","display_name":"Bellman equation","level":2,"score":0.323199987411499},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.32190001010894775},{"id":"https://openalex.org/C96250715","wikidata":"https://www.wikidata.org/wiki/Q965330","display_name":"Estimation","level":2,"score":0.3140999972820282},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.2953000068664551},{"id":"https://openalex.org/C196340769","wikidata":"https://www.wikidata.org/wiki/Q7698910","display_name":"Temporal difference learning","level":3,"score":0.2921999990940094},{"id":"https://openalex.org/C2778348673","wikidata":"https://www.wikidata.org/wiki/Q739302","display_name":"Production (economics)","level":2,"score":0.28060001134872437},{"id":"https://openalex.org/C52740198","wikidata":"https://www.wikidata.org/wiki/Q1539564","display_name":"Importance sampling","level":3,"score":0.28049999475479126},{"id":"https://openalex.org/C2776291640","wikidata":"https://www.wikidata.org/wiki/Q2912517","display_name":"Value (mathematics)","level":2,"score":0.26820001006126404},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.25870001316070557},{"id":"https://openalex.org/C14036430","wikidata":"https://www.wikidata.org/wiki/Q3736076","display_name":"Function (biology)","level":2,"score":0.2565999925136566},{"id":"https://openalex.org/C43126263","wikidata":"https://www.wikidata.org/wiki/Q128751","display_name":"Source code","level":2,"score":0.25279998779296875},{"id":"https://openalex.org/C76474335","wikidata":"https://www.wikidata.org/wiki/Q215551","display_name":"Comparative advantage","level":2,"score":0.25040000677108765}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.28963","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.28963","pdf_url":"https://arxiv.org/pdf/2609.28963","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2609.28963","doi":"https://doi.org/10.48550/arxiv.2609.28963","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.28963","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2609.28963","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.28963","pdf_url":"https://arxiv.org/pdf/2609.28963","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G6360183275","display_name":null,"funder_award_id":"22DZ2229005","funder_id":"https://openalex.org/F4320321885","funder_display_name":"Science and Technology Commission of Shanghai Municipality"}],"funders":[{"id":"https://openalex.org/F4320316083","display_name":"Tencent","ror":"https://ror.org/00hhjss72"},{"id":"https://openalex.org/F4320321885","display_name":"Science and Technology Commission of Shanghai Municipality","ror":"https://ror.org/03kt66j61"},{"id":"https://openalex.org/F4320322999","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7214358361.pdf","grobid_xml":"https://content.openalex.org/works/W7214358361.grobid-xml"},"referenced_works_count":29,"referenced_works":["https://openalex.org/W2736601468","https://openalex.org/W2978455699","https://openalex.org/W4223908421","https://openalex.org/W4391631327","https://openalex.org/W4402672014","https://openalex.org/W4404782689","https://openalex.org/W4407174758","https://openalex.org/W4414281281","https://openalex.org/W4414971194","https://openalex.org/W4415068586","https://openalex.org/W4416035997","https://openalex.org/W4416061003","https://openalex.org/W4416190659","https://openalex.org/W4416679360","https://openalex.org/W4416774044","https://openalex.org/W4417092508","https://openalex.org/W4417422257","https://openalex.org/W6891936057","https://openalex.org/W6929370958","https://openalex.org/W7079506127","https://openalex.org/W7081576550","https://openalex.org/W7102388436","https://openalex.org/W7127380112","https://openalex.org/W7134953399","https://openalex.org/W7140115172","https://openalex.org/W7154202256","https://openalex.org/W7160927012","https://openalex.org/W7165749951","https://openalex.org/W7196953391"],"related_works":["https://openalex.org/W7162494674","https://openalex.org/W7165655450","https://openalex.org/W7214775583","https://openalex.org/W7167617994","https://openalex.org/W7154202256","https://openalex.org/W7131883547","https://openalex.org/W7103150324","https://openalex.org/W7207819884","https://openalex.org/W4417093501","https://openalex.org/W7140115172"],"abstract_inverted_index":{"Group-based":[0],"reinforcement":[1],"learning":[2],"(RL)":[3],"methods,":[4],"such":[5],"as":[6],"GRPO":[7,228],"and":[8,19,162,229],"its":[9,80],"variants,":[10],"have":[11],"become":[12],"a":[13,102,128,137,150,218],"leading":[14],"paradigm":[15],"for":[16,208],"training":[17],"reasoning":[18],"agentic":[20,222,235],"large":[21],"language":[22],"models":[23],"(LLMs).":[24],"While":[25],"their":[26],"group-normalized":[27],"advantage":[28,81,177,183,194],"estimation":[29,82,109,214],"is":[30,124],"reliable":[31],"at":[32,40,242],"the":[33,41,53,66,87,90,98,107,160,169,174,181,190,205,210],"response":[34],"level,":[35,43],"it":[36],"becomes":[37],"systematically":[38],"biased":[39],"step":[42],"since":[44],"coarse-grained":[45],"trajectory-level":[46],"advantages":[47],"are":[48],"hard":[49],"to":[50,86,110,165,180,204,233],"accurately":[51],"reflect":[52],"contribution":[54],"of":[55,93,192,212,220],"individual":[56],"steps":[57],"(i.e,":[58],"failed":[59],"trajectories":[60,148],"may":[61],"contain":[62],"valuable":[63],"steps).":[64],"Revisiting":[65],"foundational":[67],"RL":[68,236],"definition,":[69],"we":[70,135,196],"notice":[71],"that":[72,144],"GRPO's":[73],"success":[74],"on":[75,127,159],"single-turn":[76],"tasks":[77],"stems":[78],"from":[79,97,119],"strategy,":[83],"which":[84,123,201],"adheres":[85,179],"basic":[88,182],"definition:":[89],"mean":[91],"reward":[92],"multiple":[94,117],"actions":[95,118],"sampled":[96],"same":[99],"state":[100],"constitutes":[101],"credible":[103],"state-value":[104,213],"estimate.":[105],"Extending":[106],"faithful":[108],"step-level":[111,176,193],"would":[112],"in":[113,185],"principle":[114],"demand":[115],"sampling":[116],"each":[120,166],"intermediate":[121],"state,":[122],"too":[125],"costly":[126],"per-state":[129],"basis.":[130],"To":[131,187],"mitigate":[132],"this":[133],"issue,":[134],"propose":[136,198],"Graph-based":[138],"Faithful":[139],"sTep-level":[140],"credit-assignment":[141],"framework":[142],"(GRAFT)":[143],"grafts":[145],"all":[146],"rollout":[147],"into":[149],"trajectory":[151,206],"graph,":[152,161],"recovering":[153],"node":[154,170],"state-values":[155],"via":[156],"Bellman":[157],"iteration":[158],"assigning":[163],"credit":[164],"edge":[167],"by":[168],"value":[171],"difference.":[172],"Theoretically,":[173],"estimated":[175],"faithfully":[178],"definition":[184],"RL.":[186],"further":[188,197],"ensure":[189],"reliability":[191],"estimation,":[195],"Graph":[199],"GAE,":[200],"extends":[202],"GAE":[203],"graph":[207],"reducing":[209],"impact":[211],"bias.":[215],"Experiments":[216],"across":[217],"range":[219],"multi-turn":[221],"benchmarks":[223],"show":[224],"consistent":[225],"gains":[226],"over":[227],"superior":[230],"performance":[231],"compared":[232],"recent":[234],"algorithms.":[237],"Code":[238],"will":[239],"be":[240],"available":[241],"https://github.com/xcyao00/GRAFT.":[243]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-09-26T00:00:00"}