{"paper":{"title":"Vidar: Embodied Video Diffusion Model for Generalist Manipulation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A video diffusion model pre-trained on internet-scale data and 750K robot trajectories adapts to new robot embodiments with only 20 minutes of demonstrations.","cross_cats":["cs.AI","cs.CV","cs.RO"],"primary_cat":"cs.LG","authors_text":"Chendong Xiang, Guodong Liu, Hang Su, Hengkai Tan, Jun Zhu, Shuhe Huang, Xinyi Mao, Yao Feng","submitted_at":"2025-07-17T08:31:55Z","abstract_excerpt":"Scaling general-purpose manipulation to new robot embodiments remains challenging: each platform typically needs large, homogeneous demonstrations, and end-to-end pixel-to-action pipelines may degenerate under background and viewpoint shifts. Based on previous advances in video-based robot control, we present Vidar, consisting of an embodied video diffusion model as the generalizable prior and a masked inverse dynamics model (MIDM) as the adapter. We leverage a video diffusion model pre-trained at Internet scale, and further continuously pre-train it for the embodied domain using 750K multi-vi"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"With only 20 minutes of human demonstrations on an unseen robot (1% of typical data), Vidar outperforms state-of-the-art baselines and generalizes to unseen tasks, backgrounds, and camera layouts.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That continuous pre-training of an internet-scale video diffusion model on 750K trajectories from only three robot platforms produces a sufficiently general visual-dynamics prior that can be grounded to arbitrary new embodiments via a lightweight masked inverse dynamics adapter.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Vidar shows that a video diffusion prior continuously pre-trained on 750K multi-view robot trajectories plus a label-free masked inverse dynamics adapter can generalize manipulation to new robot embodiments with 1% of typical demonstration data.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A video diffusion model pre-trained on internet-scale data and 750K robot trajectories adapts to new robot embodiments with only 20 minutes of demonstrations.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"c0533665f2f93f4230fdbfa435f80abef7506b75c76ee93708999befbbe66809"},"source":{"id":"2507.12898","kind":"arxiv","version":4},"verdict":{"id":"47d47346-9e8c-4074-9134-099faeccb479","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T09:50:48.324061Z","strongest_claim":"With only 20 minutes of human demonstrations on an unseen robot (1% of typical data), Vidar outperforms state-of-the-art baselines and generalizes to unseen tasks, backgrounds, and camera layouts.","one_line_summary":"Vidar shows that a video diffusion prior continuously pre-trained on 750K multi-view robot trajectories plus a label-free masked inverse dynamics adapter can generalize manipulation to new robot embodiments with 1% of typical demonstration data.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That continuous pre-training of an internet-scale video diffusion model on 750K trajectories from only three robot platforms produces a sufficiently general visual-dynamics prior that can be grounded to arbitrary new embodiments via a lightweight masked inverse dynamics adapter.","pith_extraction_headline":"A video diffusion model pre-trained on internet-scale data and 750K robot trajectories adapts to new robot embodiments with only 20 minutes of demonstrations."},"references":{"count":46,"sample":[{"doi":"","year":2024,"title":"Open X-Embodiment: Robotic Learning Datasets and RT-X Models : Open X-Embodiment Collaboration","work_id":"5e1215f6-254f-4463-b555-ef06efc0e71d","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"OpenVLA: An Open-Source Vision-Language-Action Model","work_id":"c58a80f3-556e-4070-8e8e-8969c4f0a263","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"RDT-1B: a Diffusion Foundation Model for Bimanual Manipulation","work_id":"12319725-bc7d-4c32-a229-ad270a7460bc","ref_index":3,"cited_arxiv_id":"2410.07864","is_internal_anchor":true},{"doi":"","year":2023,"title":"Crossformer: Transformer Utilizing Cross-Dimension Depen- dency for Multivariate Time Series Forecasting","work_id":"fbe803e5-9aa0-4f94-8c82-4c3eab7b9890","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"$\\pi_{0.5}$: a Vision-Language-Action Model with Open-World Generalization","work_id":"d1ad7304-d09a-49bc-809e-846439f6aff9","ref_index":5,"cited_arxiv_id":"2504.16054","is_internal_anchor":true}],"resolved_work":46,"snapshot_sha256":"45cd919227676cd242ef839aeb74f4319d3c288045301f80474c8cf998cff2fc","internal_anchors":20},"formal_canon":{"evidence_count":1,"snapshot_sha256":"836e78ec6461bff8a138f46abccd699d4b599eb8df19712fe6821ad38c1f73dc"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}