{"paper":{"title":"MVP-LAM: Learning Action-Centric Latent Action via Cross-Viewpoint Reconstruction","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Cross-viewpoint reconstruction trains latent actions to capture underlying robot actions rather than camera-specific details.","cross_cats":["cs.CV"],"primary_cat":"cs.RO","authors_text":"Dohyeok Lee, Jin Woo Koo, Jung Min Lee, Jungwoo Lee, Li Zhao, Sangwoo Hong, Seokhun Ju, Taehyun Cho","submitted_at":"2026-02-03T15:51:25Z","abstract_excerpt":"Latent actions learned from diverse human videos serve as pseudo-labels for vision-language-action (VLA) pretraining, but provide effective supervision only if they remain informative about the underlying ground-truth actions. For effective supervision, latent actions should contain information about the underlying actions even though they are inaccessible. We propose Multi-ViewPoint Latent Action Moel (MVP-LAM), which learns latent actions that are highly informative about ground-truth actions from multi-view videos. MVP-LAM trains latent actions with a cross-viewpoint reconstruction objectiv"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"MVP-LAM produces more action-centric latent actions, achieving higher mutual information with ground-truth actions and improved action prediction, including under out-of-distribution evaluation. Finally, pretraining VLAs with MVP-LAM latent actions improves downstream manipulation performance on various benchmarks.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That forcing a latent action from one viewpoint to explain the future in another viewpoint will make the latent action contain information about the underlying ground-truth actions rather than viewpoint-specific cues.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"MVP-LAM learns action-centric latent actions from multi-view videos via cross-viewpoint reconstruction, yielding higher mutual information with ground-truth actions and improved downstream VLA manipulation performance.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Cross-viewpoint reconstruction trains latent actions to capture underlying robot actions rather than camera-specific details.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"e9f85e4e3c684adf15bd89c51595407fc5614ba562c36a381fa0d59d78391f3d"},"source":{"id":"2602.03668","kind":"arxiv","version":3},"verdict":{"id":"4791e08e-34dd-42c2-a1ad-5cfcd5a82804","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T07:47:36.715161Z","strongest_claim":"MVP-LAM produces more action-centric latent actions, achieving higher mutual information with ground-truth actions and improved action prediction, including under out-of-distribution evaluation. Finally, pretraining VLAs with MVP-LAM latent actions improves downstream manipulation performance on various benchmarks.","one_line_summary":"MVP-LAM learns action-centric latent actions from multi-view videos via cross-viewpoint reconstruction, yielding higher mutual information with ground-truth actions and improved downstream VLA manipulation performance.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That forcing a latent action from one viewpoint to explain the future in another viewpoint will make the latent action contain information about the underlying ground-truth actions rather than viewpoint-specific cues.","pith_extraction_headline":"Cross-viewpoint reconstruction trains latent actions to capture underlying robot actions rather than camera-specific details."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2602.03668/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"73afbef700494a2e36d6b5812d9ab7e286420d4f39f91637ba681de9fddad6f1"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}