{"id":"https://openalex.org/W7139960851","doi":"https://doi.org/10.48550/arxiv.2603.19233","title":"Not All Features Are Created Equal: A Mechanistic Study of Vision-Language-Action Models","display_name":"Not All Features Are Created Equal: A Mechanistic Study of Vision-Language-Action Models","publication_year":2026,"publication_date":"2026-03-19","ids":{"openalex":"https://openalex.org/W7139960851","doi":"https://doi.org/10.48550/arxiv.2603.19233"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.19233","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.19233","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":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2603.19233","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5003911371","display_name":"Bryce N. Grant","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Grant, Bryce","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130231981","display_name":"Xijia Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Xijia","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5130232069","display_name":"Peng Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Peng","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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.6682000160217285,"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":0.6682000160217285,"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/T11431","display_name":"Action Observation and Synchronization","score":0.09719999879598618,"subfield":{"id":"https://openalex.org/subfields/3207","display_name":"Social Psychology"},"field":{"id":"https://openalex.org/fields/32","display_name":"Psychology"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10653","display_name":"Robot Manipulation and Learning","score":0.03539999946951866,"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/context","display_name":"Context (archaeology)","score":0.6301000118255615},{"id":"https://openalex.org/keywords/task","display_name":"Task (project management)","score":0.6071000099182129},{"id":"https://openalex.org/keywords/semantics","display_name":"Semantics (computer science)","score":0.5533999800682068},{"id":"https://openalex.org/keywords/identification","display_name":"Identification (biology)","score":0.5404999852180481},{"id":"https://openalex.org/keywords/action","display_name":"Action (physics)","score":0.4797999858856201},{"id":"https://openalex.org/keywords/representation","display_name":"Representation (politics)","score":0.4596000015735626},{"id":"https://openalex.org/keywords/encode","display_name":"ENCODE","score":0.45559999346733093},{"id":"https://openalex.org/keywords/fidelity","display_name":"Fidelity","score":0.4311000108718872},{"id":"https://openalex.org/keywords/motor-planning","display_name":"Motor planning","score":0.4059999883174896}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6518999934196472},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.6301000118255615},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.6071000099182129},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5730000138282776},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.5533999800682068},{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.5404999852180481},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.4797999858856201},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.4596000015735626},{"id":"https://openalex.org/C66746571","wikidata":"https://www.wikidata.org/wiki/Q1134833","display_name":"ENCODE","level":3,"score":0.45559999346733093},{"id":"https://openalex.org/C2776459999","wikidata":"https://www.wikidata.org/wiki/Q2119376","display_name":"Fidelity","level":2,"score":0.4311000108718872},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.4115000069141388},{"id":"https://openalex.org/C2780576329","wikidata":"https://www.wikidata.org/wiki/Q39045793","display_name":"Motor planning","level":2,"score":0.4059999883174896},{"id":"https://openalex.org/C136197465","wikidata":"https://www.wikidata.org/wiki/Q1729295","display_name":"Variety (cybernetics)","level":2,"score":0.4056999981403351},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4016999900341034},{"id":"https://openalex.org/C21200559","wikidata":"https://www.wikidata.org/wiki/Q7451068","display_name":"Sensitivity (control systems)","level":2,"score":0.3953999876976013},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.3707999885082245},{"id":"https://openalex.org/C90509273","wikidata":"https://www.wikidata.org/wiki/Q11012","display_name":"Robot","level":2,"score":0.3587000072002411},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.3237999975681305},{"id":"https://openalex.org/C66024118","wikidata":"https://www.wikidata.org/wiki/Q1122506","display_name":"Computational model","level":2,"score":0.31520000100135803},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.31380000710487366},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.2935999929904938},{"id":"https://openalex.org/C25343380","wikidata":"https://www.wikidata.org/wiki/Q277521","display_name":"Relation (database)","level":2,"score":0.289900004863739},{"id":"https://openalex.org/C32834561","wikidata":"https://www.wikidata.org/wiki/Q660730","display_name":"Subspace topology","level":2,"score":0.2858000099658966},{"id":"https://openalex.org/C137813230","wikidata":"https://www.wikidata.org/wiki/Q2996165","display_name":"Motor control","level":2,"score":0.28299999237060547},{"id":"https://openalex.org/C49937458","wikidata":"https://www.wikidata.org/wiki/Q2599292","display_name":"Probabilistic logic","level":2,"score":0.2815999984741211},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.2745000123977661},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.2702000141143799},{"id":"https://openalex.org/C178253425","wikidata":"https://www.wikidata.org/wiki/Q162668","display_name":"Visual perception","level":3,"score":0.2669999897480011},{"id":"https://openalex.org/C148220186","wikidata":"https://www.wikidata.org/wiki/Q7111912","display_name":"Outcome (game theory)","level":2,"score":0.2590000033378601},{"id":"https://openalex.org/C203005215","wikidata":"https://www.wikidata.org/wiki/Q79798","display_name":"Machine translation","level":2,"score":0.25699999928474426}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.19233","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.19233","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":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2603.19233","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.19233","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":"Preprint"},"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":[2,36],"combine":[3],"perception,":[4],"language,":[5],"and":[6,31,160,192],"motor":[7,85,145],"control":[8],"in":[9],"a":[10,119],"single":[11],"architecture,":[12],"yet":[13],"how":[14],"they":[15],"translate":[16],"multimodal":[17],"inputs":[18],"into":[19,59],"actions":[20],"remains":[21],"poorly":[22],"understood.":[23],"We":[24,205],"apply":[25],"activation":[26,167],"injection,":[27],"sparse":[28],"autoencoders":[29],"(SAEs),":[30],"linear":[32],"probes":[33],"to":[34,88],"six":[35,218],"spanning":[37,197],"80M--7B":[38],"parameters":[39],"across":[40,53,216],"394,000+":[41],"rollout":[42],"episodes":[43,61,76],"on":[44,99,177,184],"four":[45],"benchmarks.":[46],"The":[47],"visual":[48,106],"pathway":[49],"dominates":[50],"action":[51,175],"generation":[52],"all":[54,135,217],"architectures:":[55],"injecting":[56],"baseline":[57],"activations":[58],"null-prompt":[60],"recovers":[62,188],"near-identical":[63],"behavior,":[64],"while":[65,147],"cross-task":[66],"injection":[67,162],"steers":[68],"robots":[69],"toward":[70],"source-task":[71],"positions":[72],"(99.8\\%":[73],"of":[74,202,213],"X-VLA":[75],"align":[77],"with":[78],"the":[79,110],"source":[80],"trajectory),":[81],"exposing":[82],"spatially":[83],"bound":[84],"programs":[86,146],"tied":[87],"scene":[89],"coordinates":[90],"rather":[91],"than":[92],"abstract":[93],"task":[94,100],"representations.":[95],"Language":[96],"sensitivity":[97,196],"depends":[98],"structure,":[101],"not":[102],"model":[103],"design:":[104],"when":[105,115],"context":[107],"uniquely":[108],"specifies":[109],"task,":[111],"language":[112,121],"is":[113,172],"ignored;":[114],"multiple":[116],"goals":[117],"share":[118],"scene,":[120],"becomes":[122],"essential":[123,173],"(X-VLA":[124],"\\texttt{libero\\_goal}:":[125],"94\\%$\\to$10\\%":[126],"under":[127],"wrong":[128],"prompts":[129],"vs.\\":[130],"\\texttt{libero\\_object}:":[131],"60--100\\%":[132],"regardless).":[133],"In":[134],"three":[136],"multi-pathway":[137],"architectures":[138],"(\\pizhalf{},":[139],"SmolVLA,":[140],"GR00T),":[141],"expert":[142,158],"pathways":[143,149],"encode":[144,150],"VLM":[148],"goal":[151],"semantics":[152],"($2\\times$":[153],"greater":[154],"behavioral":[155],"displacement":[156],"from":[157],"injection),":[159],"subspace":[161],"confirms":[163],"these":[164],"occupy":[165],"separable":[166],"subspaces.":[168],"Per-token":[169],"SAE":[170],"processing":[171],"for":[174,210],"fidelity":[176,183],"most":[178],"architectures,":[179],"though":[180],"mean-pooling":[181],"improves":[182],"X-VLA.":[185],"Contrastive":[186],"identification":[187],"82+":[189],"manipulation":[190],"concepts,":[191],"causal":[193],"ablation":[194],"reveals":[195],"28--92\\%":[198],"zero-effect":[199],"rates":[200],"independent":[201],"representation":[203],"width.":[204],"release":[206],"\\textbf{Action":[207],"Atlas}":[208],"(https://action-atlas.com)":[209],"interactive":[211],"exploration":[212],"VLA":[214],"representations":[215],"models.":[219]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-03-21T00:00:00"}