{"paper":{"title":"Any-point Trajectory Modeling for Policy Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Pre-training a model to predict future trajectories of arbitrary points in videos supplies control guidance that lets robots learn policies from minimal action-labeled data.","cross_cats":["cs.CV"],"primary_cat":"cs.RO","authors_text":"Chuan Wen, John So, Kai Chen, Pieter Abbeel, Qi Dou, Xingyu Lin, Yang Gao","submitted_at":"2023-12-28T23:34:43Z","abstract_excerpt":"Learning from demonstration is a powerful method for teaching robots new skills, and having more demonstration data often improves policy learning. However, the high cost of collecting demonstration data is a significant bottleneck. Videos, as a rich data source, contain knowledge of behaviors, physics, and semantics, but extracting control-specific information from them is challenging due to the lack of action labels. In this work, we introduce a novel framework, Any-point Trajectory Modeling (ATM), that utilizes video demonstrations by pre-training a trajectory model to predict future trajec"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across over 130 language-conditioned tasks we evaluated in both simulation and the real world, ATM outperforms strong video pre-training baselines by 80% on average.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That predicted trajectories of arbitrary points supply sufficiently accurate and transferable control guidance to enable robust policy learning from only minimal action-labeled data.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"ATM pre-trains models to predict trajectories of any points in videos, then uses those predictions to learn strong visuomotor policies from minimal action labels, beating baselines by 80% on 130+ tasks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Pre-training a model to predict future trajectories of arbitrary points in videos supplies control guidance that lets robots learn policies from minimal action-labeled data.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"2a7cb39692e3d4d76f56771bb608dfd2dbe79a6af1017b9004ea210571b0f3c2"},"source":{"id":"2401.00025","kind":"arxiv","version":3},"verdict":{"id":"60a17f23-8822-4180-82cb-177037712d97","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T23:27:58.879801Z","strongest_claim":"Across over 130 language-conditioned tasks we evaluated in both simulation and the real world, ATM outperforms strong video pre-training baselines by 80% on average.","one_line_summary":"ATM pre-trains models to predict trajectories of any points in videos, then uses those predictions to learn strong visuomotor policies from minimal action labels, beating baselines by 80% on 130+ tasks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That predicted trajectories of arbitrary points supply sufficiently accurate and transferable control guidance to enable robust policy learning from only minimal action-labeled data.","pith_extraction_headline":"Pre-training a model to predict future trajectories of arbitrary points in videos supplies control guidance that lets robots learn policies from minimal action-labeled data."},"references":{"count":56,"sample":[{"doi":"","year":2007,"title":"Trajectory- tracking and path-following of underactuated au- tonomous vehicles with parametric modeling uncertainty","work_id":"5d129caa-4470-446b-846b-2f93e56fa144","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Affordances from human videos 417 as a versatile rep- resentation for robotics","work_id":"bb92ac12-300b-4596-b676-cdc3976e29c1","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Video pretraining (vpt): Learning to act by watching unlabeled online videos","work_id":"008a87ff-6368-4639-b2a4-91059d0f90e5","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Zero-shot robot manipu- lation from passive human videos","work_id":"69f1ea0a-5458-4344-9835-4e29765c78e1","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Zero-Shot Robotic Manipulation with Pretrained Image-Editing Diffusion Models","work_id":"954b4359-f4ed-4c73-ae5b-f75d486b6fc8","ref_index":5,"cited_arxiv_id":"2310.10639","is_internal_anchor":true}],"resolved_work":56,"snapshot_sha256":"938093790cb94781782dbeabc28d34840063a6deb7c9842712caca02fd0269be","internal_anchors":6},"formal_canon":{"evidence_count":2,"snapshot_sha256":"9187ea1df91f2e5a6bf976650ba3caeb7965e49d028fa35436ffdce100d4291b"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}