{"id":"https://openalex.org/W7162135431","doi":"https://doi.org/10.48550/arxiv.2605.22177","title":"Maestro: Reinforcement Learning to Orchestrate Hierarchical Model-Skill Ensembles","display_name":"Maestro: Reinforcement Learning to Orchestrate Hierarchical Model-Skill Ensembles","publication_year":2026,"publication_date":"2026-05-21","ids":{"openalex":"https://openalex.org/W7162135431","doi":"https://doi.org/10.48550/arxiv.2605.22177"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.22177","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.22177","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.22177","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5136777023","display_name":"Jinyang Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Jinyang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5122335801","display_name":"Guocheng Zhai","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhai, Guocheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136746000","display_name":"Ruihan Jin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jin, Ruihan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136809441","display_name":"Yuhao Shen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shen, Yuhao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136754050","display_name":"Zhengxi Lu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lu, Zhengxi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136730624","display_name":"Fan Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Fan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136786542","display_name":"Haoran Luo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Luo, Haoran","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136775574","display_name":"Zheng Lian","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lian, Zheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136814902","display_name":"Zhengqi Wen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wen, Zhengqi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5136737405","display_name":"Jianhua Tao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tao, Jianhua","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":10,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.32440000772476196,"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":0.32440000772476196,"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/T10028","display_name":"Topic Modeling","score":0.2451999932527542,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.05169999971985817,"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":0.7038000226020813},{"id":"https://openalex.org/keywords/exploit","display_name":"Exploit","score":0.654699981212616},{"id":"https://openalex.org/keywords/modular-design","display_name":"Modular design","score":0.6204000115394592},{"id":"https://openalex.org/keywords/process","display_name":"Process (computing)","score":0.5913000106811523},{"id":"https://openalex.org/keywords/code","display_name":"Code (set theory)","score":0.5022000074386597},{"id":"https://openalex.org/keywords/limiting","display_name":"Limiting","score":0.499099999666214},{"id":"https://openalex.org/keywords/relevance","display_name":"Relevance (law)","score":0.43560001254081726},{"id":"https://openalex.org/keywords/orchestration","display_name":"Orchestration","score":0.43389999866485596}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.771399974822998},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.7038000226020813},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.654699981212616},{"id":"https://openalex.org/C101468663","wikidata":"https://www.wikidata.org/wiki/Q1620158","display_name":"Modular design","level":2,"score":0.6204000115394592},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.5913000106811523},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5231000185012817},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.5022000074386597},{"id":"https://openalex.org/C188198153","wikidata":"https://www.wikidata.org/wiki/Q1613840","display_name":"Limiting","level":2,"score":0.499099999666214},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4616999924182892},{"id":"https://openalex.org/C158154518","wikidata":"https://www.wikidata.org/wiki/Q7310970","display_name":"Relevance (law)","level":2,"score":0.43560001254081726},{"id":"https://openalex.org/C199168358","wikidata":"https://www.wikidata.org/wiki/Q3367000","display_name":"Orchestration","level":3,"score":0.43389999866485596},{"id":"https://openalex.org/C190839683","wikidata":"https://www.wikidata.org/wiki/Q2448197","display_name":"Train","level":2,"score":0.42750000953674316},{"id":"https://openalex.org/C113843644","wikidata":"https://www.wikidata.org/wiki/Q901882","display_name":"Interface (matter)","level":4,"score":0.39719998836517334},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.3772999942302704},{"id":"https://openalex.org/C190812933","wikidata":"https://www.wikidata.org/wiki/Q28923","display_name":"Chart","level":2,"score":0.31450000405311584},{"id":"https://openalex.org/C189430467","wikidata":"https://www.wikidata.org/wiki/Q7293293","display_name":"Ranking (information retrieval)","level":2,"score":0.29170000553131104},{"id":"https://openalex.org/C22367795","wikidata":"https://www.wikidata.org/wiki/Q7625208","display_name":"Structured prediction","level":2,"score":0.2874999940395355},{"id":"https://openalex.org/C110332635","wikidata":"https://www.wikidata.org/wiki/Q629498","display_name":"Genetic programming","level":2,"score":0.2872999906539917},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.2806999981403351},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.2777999937534332},{"id":"https://openalex.org/C43126263","wikidata":"https://www.wikidata.org/wiki/Q128751","display_name":"Source code","level":2,"score":0.2759000062942505},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.27320000529289246},{"id":"https://openalex.org/C77967617","wikidata":"https://www.wikidata.org/wiki/Q4677561","display_name":"Active learning (machine learning)","level":2,"score":0.27000001072883606}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.22177","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.22177","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"raw_source_name":null,"raw_type":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.22177","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.22177","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":"article"},"sustainable_development_goals":[{"display_name":"Peace, Justice and strong institutions","id":"https://metadata.un.org/sdg/16","score":0.4693875014781952}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"The":[0,148,236],"proliferation":[1],"of":[2,57,120,187],"large":[3],"language":[4],"models":[5,58,123,204],"(LLMs)":[6],"and":[7,25,59,124,144,174,193,205],"modular":[8],"skills":[9,206],"has":[10],"endowed":[11],"autonomous":[12],"agents":[13],"with":[14,30,212,233],"increasingly":[15],"powerful":[16],"capabilities.":[17],"Existing":[18],"frameworks":[19,50],"typically":[20],"rely":[21],"on":[22,65,219],"monolithic":[23],"LLMs":[24,41],"fixed":[26],"logic":[27],"to":[28,36,52,116,134,142,146,202],"interface":[29],"these":[31],"skills.":[32],"This":[33],"gives":[34],"rise":[35],"a":[37,81,93,98,108,113,125,179,216],"critical":[38],"bottleneck:":[39],"different":[40],"offer":[42],"distinct":[43],"advantages":[44],"across":[45,162],"diverse":[46],"domains,":[47],"yet":[48],"current":[49],"fail":[51],"exploit":[53],"the":[54,197,210],"complementary":[55],"strengths":[56],"skills,":[60],"thereby":[61],"limiting":[62],"their":[63],"performance":[64],"downstream":[66],"tasks.":[67],"In":[68],"this":[69],"paper,":[70],"we":[71],"present":[72],"Maestro":[73,111,161,182,227],"(Multimodal":[74],"Agent":[75],"for":[76],"Expert-Skill":[77],"Targeted":[78],"Reinforced":[79],"Orchestration),":[80],"Reinforcement":[82],"Learning":[83],"(RL)-driven":[84],"orchestration":[85],"framework":[86],"that":[87],"reframes":[88],"heterogeneous":[89],"multimodal":[90,165],"tasks":[91],"as":[92],"sequential":[94],"decision-making":[95],"process":[96],"over":[97],"hierarchical":[99],"model-skill":[100,140],"registry.":[101],"Rather":[102],"than":[103],"consolidating":[104],"all":[105,224],"knowledge":[106],"into":[107],"single":[109],"model,":[110],"trains":[112],"lightweight":[114],"policy":[115,149,200],"dynamically":[117],"compose":[118],"ensembles":[119],"frozen":[121],"expert":[122],"two-tier":[126],"skill":[127],"library,":[128],"deciding":[129],"at":[130,241],"each":[131],"step":[132],"whether":[133],"invoke":[135],"an":[136,184],"external":[137],"expert,":[138],"which":[139],"pair":[141],"select,":[143],"when":[145],"terminate.":[147],"is":[150,239],"optimized":[151],"via":[152],"outcome-based":[153],"RL,":[154],"requiring":[155],"no":[156],"step-level":[157],"supervision.":[158],"We":[159],"evaluate":[160],"ten":[163],"representative":[164],"benchmarks":[166],"spanning":[167],"mathematical":[168],"reasoning,":[169],"chart":[170],"understanding,":[171],"high-resolution":[172],"perception,":[173],"domain-specific":[175],"analysis.":[176],"With":[177],"only":[178],"4B":[180],"orchestrator,":[181],"achieves":[183],"average":[185,218],"accuracy":[186],"70.1%,":[188],"surpassing":[189],"both":[190],"GPT-5":[191],"(69.3%)":[192],"Gemini-2.5-Pro":[194],"(68.7%).":[195],"Crucially,":[196],"learned":[198],"coordination":[199],"generalizes":[201],"unseen":[203],"without":[207],"retraining:":[208],"augmenting":[209],"registry":[211],"out-of-domain":[213],"experts":[214],"yields":[215],"59.5%":[217],"four":[220],"challenging":[221],"benchmarks,":[222],"outperforming":[223],"closed-source":[225],"baselines.":[226],"further":[228],"maintains":[229],"high":[230],"computational":[231],"efficiency":[232],"low":[234],"latency.":[235],"source":[237],"code":[238],"available":[240],"https://github.com/jinyangwu/Maestro.":[242]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2026-05-23T00:00:00"}