{"paper":{"title":"3D Diffusion Policy: Generalizable Visuomotor Policy Learning via Simple 3D Representations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A compact 3D point-cloud representation lets diffusion policies learn precise robot manipulation from only ten demonstrations.","cross_cats":["cs.CV","cs.LG"],"primary_cat":"cs.RO","authors_text":"Chenyuan Hu, Gu Zhang, Huazhe Xu, Kangning Zhang, Muhan Wang, Yanjie Ze","submitted_at":"2024-03-06T18:58:49Z","abstract_excerpt":"Imitation learning provides an efficient way to teach robots dexterous skills; however, learning complex skills robustly and generalizablely usually consumes large amounts of human demonstrations. To tackle this challenging problem, we present 3D Diffusion Policy (DP3), a novel visual imitation learning approach that incorporates the power of 3D visual representations into diffusion policies, a class of conditional action generative models. The core design of DP3 is the utilization of a compact 3D visual representation, extracted from sparse point clouds with an efficient point encoder. In our"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"DP3 successfully handles most tasks with just 10 demonstrations and surpasses baselines with a 24.2% relative improvement. In 4 real robot tasks, DP3 demonstrates precise control with a high success rate of 85%, given only 40 demonstrations of each task, and shows excellent generalization abilities in diverse aspects, including space, viewpoint, appearance, and instance.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the compact 3D visual representation extracted from sparse point clouds is sufficient to capture all task-relevant geometry and generalizes across the reported variations without requiring additional modalities or task-specific tuning.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"DP3 uses compact 3D representations from sparse point clouds inside diffusion policies to learn generalizable visuomotor skills from few demonstrations, reporting 24% gains in simulation and 85% success on real robots.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A compact 3D point-cloud representation lets diffusion policies learn precise robot manipulation from only ten demonstrations.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"ca4b7c96669eec40f6437a925cdf6cfb940ff9f939db2a49af05c61dc98653a8"},"source":{"id":"2403.03954","kind":"arxiv","version":7},"verdict":{"id":"f026222b-ba74-4d38-a380-7be14c7b9072","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T01:45:13.165980Z","strongest_claim":"DP3 successfully handles most tasks with just 10 demonstrations and surpasses baselines with a 24.2% relative improvement. In 4 real robot tasks, DP3 demonstrates precise control with a high success rate of 85%, given only 40 demonstrations of each task, and shows excellent generalization abilities in diverse aspects, including space, viewpoint, appearance, and instance.","one_line_summary":"DP3 uses compact 3D representations from sparse point clouds inside diffusion policies to learn generalizable visuomotor skills from few demonstrations, reporting 24% gains in simulation and 85% success on real robots.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the compact 3D visual representation extracted from sparse point clouds is sufficient to capture all task-relevant geometry and generalizes across the reported variations without requiring additional modalities or task-specific tuning.","pith_extraction_headline":"A compact 3D point-cloud representation lets diffusion policies learn precise robot manipulation from only ten demonstrations."},"references":{"count":86,"sample":[{"doi":"","year":2023,"title":"Dexterous functional grasping","work_id":"9d69378e-39e5-400e-b8d6-32281def3639","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Is Conditional Generative Modeling all you need for Decision-Making?","work_id":"dac365c0-e557-4886-9a1b-179151a66160","ref_index":2,"cited_arxiv_id":"2211.15657","is_internal_anchor":true},{"doi":"","year":2023,"title":"Dexterous imitation made easy: A learning-based framework for efficient dexterous manip- ulation","work_id":"dea89fcb-ec88-420f-98c5-98f62b27946c","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2016,"title":"Layer normalization","work_id":"ed385731-279c-494d-a83e-231e5685c903","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Dexart: Benchmarking generalizable dexterous manipu- lation with articulated objects","work_id":"29407570-41c3-421c-984f-e44dbbe81587","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":86,"snapshot_sha256":"344f58f2269bcfefd09f744cdb20674d831cb97527c15d73c90166e0af897c52","internal_anchors":3},"formal_canon":{"evidence_count":3,"snapshot_sha256":"a1db8447b7fb0c2773103f3a1554ac5f959ef739f1af420d76ce6e6df4d4ec5c"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}