{"id":"https://openalex.org/W7204185702","doi":"https://doi.org/10.48550/arxiv.2608.22631","title":"Learning Generalizable Behaviors for Terminal Agents","display_name":"Learning Generalizable Behaviors for Terminal Agents","publication_year":2026,"publication_date":"2026-08-23","ids":{"openalex":"https://openalex.org/W7204185702","doi":"https://doi.org/10.48550/arxiv.2608.22631"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2608.22631","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.22631","pdf_url":"https://arxiv.org/pdf/2608.22631","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","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2608.22631","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5148098121","display_name":"Yihang Yao","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yao, Yihang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5120340008","display_name":"Bo Pang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pang, Bo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5115647547","display_name":"Xuan Phi Nguyen","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Nguyen, Xuan Phi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5148148261","display_name":"Ding Zhao","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Ding","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5005443526","display_name":"Shafiq Joty","orcid":"https://orcid.org/0000-0002-9222-2641","observed_orcids":["https://orcid.org/0000-0002-9222-2641"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Joty, Shafiq","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5048500970","display_name":"Semih Yavuz","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yavuz, Semih","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"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":0.8163999915122986,"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":0.8163999915122986,"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":0.0333000011742115,"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/T12031","display_name":"Speech and dialogue systems","score":0.02250000089406967,"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/large-language-models","display_name":"large language models","score":0.8859999775886536},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"reinforcement learning","score":0.7960000038146973},{"id":"https://openalex.org/keywords/synthetic-training-environments","display_name":"synthetic training environments","score":0.8450000286102295},{"id":"https://openalex.org/keywords/compositional-generalization","display_name":"compositional generalization","score":0.7749999761581421},{"id":"https://openalex.org/keywords/domain-gap","display_name":"domain gap","score":0.3779999911785126}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7645999789237976},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.744700014591217},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.7172999978065491},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.5853999853134155},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5842999815940857},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.5730000138282776},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.5016999840736389},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.39719998836517334},{"id":"https://openalex.org/C36503486","wikidata":"https://www.wikidata.org/wiki/Q11235244","display_name":"Domain (mathematical analysis)","level":2,"score":0.39329999685287476},{"id":"https://openalex.org/C2779664074","wikidata":"https://www.wikidata.org/wiki/Q3518405","display_name":"Terminal (telecommunication)","level":2,"score":0.37860000133514404},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.3725999891757965},{"id":"https://openalex.org/C148220186","wikidata":"https://www.wikidata.org/wiki/Q7111912","display_name":"Outcome (game theory)","level":2,"score":0.36039999127388},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.3425000011920929},{"id":"https://openalex.org/C67203356","wikidata":"https://www.wikidata.org/wiki/Q1321905","display_name":"Reinforcement","level":2,"score":0.33739998936653137},{"id":"https://openalex.org/C136197465","wikidata":"https://www.wikidata.org/wiki/Q1729295","display_name":"Variety (cybernetics)","level":2,"score":0.32659998536109924},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.3019999861717224},{"id":"https://openalex.org/C115051666","wikidata":"https://www.wikidata.org/wiki/Q6522493","display_name":"Ranging","level":2,"score":0.3012000024318695},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.29010000824928284},{"id":"https://openalex.org/C136389625","wikidata":"https://www.wikidata.org/wiki/Q334384","display_name":"Supervised learning","level":3,"score":0.2612000107765198},{"id":"https://openalex.org/C120936955","wikidata":"https://www.wikidata.org/wiki/Q2155640","display_name":"Empirical research","level":2,"score":0.2605000138282776},{"id":"https://openalex.org/C2781316041","wikidata":"https://www.wikidata.org/wiki/Q1230584","display_name":"Diversity (politics)","level":2,"score":0.251800000667572},{"id":"https://openalex.org/C2776502983","wikidata":"https://www.wikidata.org/wiki/Q690182","display_name":"Contrast (vision)","level":2,"score":0.25040000677108765}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2608.22631","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.22631","pdf_url":"https://arxiv.org/pdf/2608.22631","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.2608.22631","doi":"https://doi.org/10.48550/arxiv.2608.22631","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.22631","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:2608.22631","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.22631","pdf_url":"https://arxiv.org/pdf/2608.22631","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":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7125695819","https://openalex.org/W7168255382","https://openalex.org/W7165627739","https://openalex.org/W7161983523","https://openalex.org/W7162785433","https://openalex.org/W7221807298","https://openalex.org/W7135215173","https://openalex.org/W7212293382","https://openalex.org/W7130534234","https://openalex.org/W7131402252"],"abstract_inverted_index":{"Terminal":[0],"agents":[1,92],"are":[2,44,144],"a":[3,25,36,49,163],"compelling":[4],"application":[5],"of":[6,73,219],"large":[7],"language":[8],"models":[9,198,235],"(LLMs),":[10],"with":[11,131,179],"the":[12,69,80,95,190,220],"potential":[13],"to":[14,62,239],"integrate":[15],"deeply":[16],"into":[17],"users'":[18],"daily":[19],"workflows.":[20],"Reinforcement":[21],"learning":[22],"(RL)":[23],"is":[24,129,146],"key":[26],"technique":[27],"for":[28,234],"improving":[29],"their":[30],"capabilities,":[31],"making":[32],"scalable":[33],"training":[34,165,222],"environments":[35,47,174],"central":[37],"challenge.":[38],"Since":[39],"public":[40],"real-user":[41],"interaction":[42],"data":[43],"scarce,":[45],"synthetic":[46,74],"provide":[48],"practical":[50],"alternative,":[51],"but":[52],"often":[53],"suffer":[54],"from":[55,106,237],"domain":[56],"gaps":[57],"and":[58,71,79,93,116,123,135,175,212,230,243],"limited":[59],"fidelity,":[60],"leading":[61],"poor":[63],"generalization.":[64],"Existing":[65],"work":[66],"mainly":[67],"scales":[68],"quantity":[70,153],"diversity":[72],"environments,":[75,223],"while":[76],"reward-signal":[77],"quality":[78,170],"mechanisms":[81],"governing":[82],"generalization":[83],"remain":[84],"under-explored.":[85],"We":[86],"study":[87],"how":[88],"RL":[89,108,213,226],"improves":[90,168,225],"terminal":[91],"propose":[94,161],"Agentic":[96],"Compositional":[97],"Generalization":[98],"hypothesis:":[99],"rather":[100],"than":[101,149,217],"teaching":[102],"new":[103],"domain-specific":[104],"skills":[105,119],"scratch,":[107],"primarily":[109],"shapes":[110],"high-level":[111],"decision-making":[112],"behaviors":[113,143],"that":[114,137,167],"compose":[115],"route":[117],"low-level":[118],"acquired":[120],"during":[121],"pre-training":[122],"supervised":[124],"fine-tuning":[125],"(SFT).":[126],"This":[127],"account":[128],"consistent":[130],"our":[132,186],"empirical":[133],"results":[134],"suggests":[136],"verifier":[138],"quality,":[139],"which":[140,142],"determines":[141],"reinforced,":[145],"more":[147],"important":[148],"simply":[150],"increasing":[151],"environment":[152],"or":[154],"diversity.":[155],"Motivated":[156],"by":[157,171,228],"this":[158,184],"insight,":[159],"we":[160],"River,":[162],"simple":[164],"recipe":[166],"reward":[169],"filtering":[172],"low-quality":[173],"augmenting":[176],"outcome":[177],"rewards":[178],"process-level":[180],"behavior":[181],"regularization.":[182],"Using":[183,215],"recipe,":[185],"RL-trained":[187,196],"agent":[188,210],"achieves":[189],"best":[191],"performance":[192],"among":[193],"evaluated":[194],"open-source":[195],"8B":[197],"across":[199,206],"four":[200],"terminal-agent":[201],"benchmarks.":[202],"River":[203,224],"also":[204],"generalizes":[205],"model":[207],"families,":[208],"scales,":[209],"harnesses,":[211],"objectives.":[214],"fewer":[216],"30%":[218,231],"TMax":[221],"gains":[227],"106%":[229],"on":[232,241],"average":[233],"ranging":[236],"2B":[238],"27B":[240],"Terminal-Bench-Lite":[242],"Terminal-Bench-v2.1,":[244],"respectively.":[245]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-08-26T00:00:00"}