{"id":"https://openalex.org/W7139930553","doi":"https://doi.org/10.48550/arxiv.2603.19183","title":"Sparse Autoencoders Reveal Interpretable and Steerable Features in VLA Models","display_name":"Sparse Autoencoders Reveal Interpretable and Steerable Features in VLA Models","publication_year":2026,"publication_date":"2026-03-19","ids":{"openalex":"https://openalex.org/W7139930553","doi":"https://doi.org/10.48550/arxiv.2603.19183"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.19183","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.19183","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.2603.19183","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5114159472","display_name":"Aiden Swann","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Swann, Aiden","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130234136","display_name":"Lachlain McGranahan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"McGranahan, Lachlain","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128240530","display_name":"Hugo Buurmeijer","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Buurmeijer, Hugo","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5047853405","display_name":"Monroe Kennedy","orcid":"https://orcid.org/0000-0002-4567-0409"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kennedy, Monroe","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5130234330","display_name":"Mac Schwager","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Schwager, Mac","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":5,"corresponding_author_ids":["https://openalex.org/A5114159472"],"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":0.37940001487731934,"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":0.37940001487731934,"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":0.12860000133514404,"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/T10653","display_name":"Robot Manipulation and Learning","score":0.11509999632835388,"subfield":{"id":"https://openalex.org/subfields/2207","display_name":"Control and Systems Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/interpretability","display_name":"Interpretability","score":0.888700008392334},{"id":"https://openalex.org/keywords/generalization","display_name":"Generalization","score":0.6611999869346619},{"id":"https://openalex.org/keywords/codebase","display_name":"Codebase","score":0.6421999931335449},{"id":"https://openalex.org/keywords/feature","display_name":"Feature (linguistics)","score":0.5447999835014343},{"id":"https://openalex.org/keywords/categorization","display_name":"Categorization","score":0.5403000116348267},{"id":"https://openalex.org/keywords/metric","display_name":"Metric (unit)","score":0.5293999910354614},{"id":"https://openalex.org/keywords/robot","display_name":"Robot","score":0.5152999758720398},{"id":"https://openalex.org/keywords/interface","display_name":"Interface (matter)","score":0.41589999198913574},{"id":"https://openalex.org/keywords/pattern-recognition","display_name":"Pattern recognition (psychology)","score":0.3833000063896179}],"concepts":[{"id":"https://openalex.org/C2781067378","wikidata":"https://www.wikidata.org/wiki/Q17027399","display_name":"Interpretability","level":2,"score":0.888700008392334},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.7296000123023987},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7172999978065491},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.6611999869346619},{"id":"https://openalex.org/C51929080","wikidata":"https://www.wikidata.org/wiki/Q2425187","display_name":"Codebase","level":3,"score":0.6421999931335449},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.5447999835014343},{"id":"https://openalex.org/C94124525","wikidata":"https://www.wikidata.org/wiki/Q912550","display_name":"Categorization","level":2,"score":0.5403000116348267},{"id":"https://openalex.org/C176217482","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Metric (unit)","level":2,"score":0.5293999910354614},{"id":"https://openalex.org/C90509273","wikidata":"https://www.wikidata.org/wiki/Q11012","display_name":"Robot","level":2,"score":0.5152999758720398},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.44850000739097595},{"id":"https://openalex.org/C113843644","wikidata":"https://www.wikidata.org/wiki/Q901882","display_name":"Interface (matter)","level":4,"score":0.41589999198913574},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.3833000063896179},{"id":"https://openalex.org/C34413123","wikidata":"https://www.wikidata.org/wiki/Q170978","display_name":"Robotics","level":3,"score":0.3801000118255615},{"id":"https://openalex.org/C104114177","wikidata":"https://www.wikidata.org/wiki/Q79782","display_name":"Motion (physics)","level":2,"score":0.37770000100135803},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.34689998626708984},{"id":"https://openalex.org/C124304363","wikidata":"https://www.wikidata.org/wiki/Q673661","display_name":"Abstraction","level":2,"score":0.3456999957561493},{"id":"https://openalex.org/C5274069","wikidata":"https://www.wikidata.org/wiki/Q2285707","display_name":"Categorical variable","level":2,"score":0.3361000120639801},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.3172999918460846},{"id":"https://openalex.org/C189645446","wikidata":"https://www.wikidata.org/wiki/Q350865","display_name":"Mirroring","level":2,"score":0.2996000051498413},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.29899999499320984},{"id":"https://openalex.org/C60692881","wikidata":"https://www.wikidata.org/wiki/Q584529","display_name":"Humanoid robot","level":3,"score":0.2858999967575073},{"id":"https://openalex.org/C136389625","wikidata":"https://www.wikidata.org/wiki/Q334384","display_name":"Supervised learning","level":3,"score":0.28209999203681946},{"id":"https://openalex.org/C171268870","wikidata":"https://www.wikidata.org/wiki/Q1486676","display_name":"GRASP","level":2,"score":0.2766000032424927},{"id":"https://openalex.org/C12426560","wikidata":"https://www.wikidata.org/wiki/Q189569","display_name":"Basis (linear algebra)","level":2,"score":0.27230000495910645},{"id":"https://openalex.org/C2777212361","wikidata":"https://www.wikidata.org/wiki/Q5127848","display_name":"Class (philosophy)","level":2,"score":0.26159998774528503},{"id":"https://openalex.org/C52622490","wikidata":"https://www.wikidata.org/wiki/Q1026626","display_name":"Feature extraction","level":2,"score":0.25209999084472656}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.19183","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.19183","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.2603.19183","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.19183","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":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Vision-Language-Action":[0],"(VLA)":[1],"models":[2,20],"have":[3],"emerged":[4],"as":[5,75],"a":[6,69,76,117,125],"promising":[7,118],"approach":[8],"for":[9,80,235],"general-purpose":[10],"robot":[11,160],"manipulation.":[12],"However,":[13,102],"their":[14,169],"generalization":[15],"is":[16,246],"inconsistent:":[17],"while":[18],"these":[19,143],"can":[21,173,189],"perform":[22],"impressively":[23],"in":[24],"some":[25,103],"settings,":[26],"fine-tuned":[27],"variants":[28],"often":[29],"fail":[30],"on":[31,60,148,202,212],"novel":[32],"objects,":[33],"scenes,":[34],"and":[35,109,113,172,178,195,232,240],"instructions.":[36],"We":[37,84,123,141,152,197,227],"apply":[38],"mechanistic":[39,185],"interpretability":[40],"techniques":[41],"to":[42,95,106,127,131],"better":[43],"understand":[44],"the":[45,65,81,87,149,183],"inner":[46],"workings":[47],"of":[48,64,90],"VLA":[49,121],"models.":[50],"To":[51],"probe":[52],"internal":[53],"representations,":[54],"we":[55],"train":[56],"Sparse":[57],"Autoencoders":[58],"(SAEs)":[59],"hidden":[61],"layer":[62],"activations":[63],"VLA.":[66],"SAEs":[67],"learn":[68,190],"sparse":[70],"dictionary":[71],"whose":[72],"features":[73,93,104,129,157,164,192],"act":[74],"compact,":[77],"interpretable":[78],"basis":[79],"model's":[82],"computation.":[83],"find":[85],"that":[86,154,187,199],"large":[88],"majority":[89],"extracted":[91],"SAE":[92,156,238],"correspond":[94,105],"memorized":[96],"sequences":[97],"from":[98],"specific":[99],"training":[100,211],"demonstrations.":[101],"interpretable,":[107],"general,":[108],"steerable":[110],"motion":[111],"primitives":[112,137],"semantic":[114,170],"properties,":[115],"offering":[116],"glimpse":[119],"toward":[120],"generalizability.":[122],"propose":[124],"metric":[126],"categorize":[128],"according":[130],"whether":[132],"they":[133],"represent":[134],"generalizable":[135,191],"transferable":[136],"or":[138,219],"episode-specific":[139],"memorization.":[140,208],"validate":[142],"findings":[144],"through":[145],"steering":[146],"experiments":[147],"LIBERO":[150],"benchmark.":[151],"show":[153],"individual":[155],"causally":[158],"influence":[159],"behavior.":[161],"Steering":[162],"general":[163,225],"induces":[165],"behaviors":[166],"consistent":[167],"with":[168],"meaning":[171],"be":[174],"applied":[175],"across":[176,193],"tasks":[177,194],"scenes.":[179,196],"This":[180],"work":[181],"provides":[182],"first":[184],"evidence":[186],"VLAs":[188],"observe":[198],"supervised":[200],"fine-tuning":[201],"small":[203],"robotics":[204],"datasets":[205,216],"disproportionately":[206],"amplifies":[207],"In":[209],"contrast,":[210],"larger,":[213],"more":[214,224],"diverse":[215],"(e.g.,":[217],"DROID)":[218],"using":[220],"knowledge":[221],"insulation":[222],"promotes":[223],"features.":[226],"provide":[228],"an":[229],"open-source":[230],"codebase":[231],"user-friendly":[233],"interface":[234],"activation":[236],"collection,":[237],"training,":[239],"feature":[241],"steering.":[242],"Our":[243],"project":[244],"page":[245],"located":[247],"at":[248],"http://drvla.github.io":[249]},"counts_by_year":[],"updated_date":"2026-03-21T06:36:02.116451","created_date":"2026-03-21T00:00:00"}