{"paper":{"title":"Universal Manipulation Interface: In-The-Wild Robot Teaching Without In-The-Wild Robots","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"UMI lets robots learn complex manipulation from portable human gripper demonstrations with zero-shot transfer to new settings and hardware.","cross_cats":[],"primary_cat":"cs.RO","authors_text":"Benjamin Burchfiel, Cheng Chi, Chuer Pan, Eric Cousineau, Russ Tedrake, Shuran Song, Siyuan Feng, Zhenjia Xu","submitted_at":"2024-02-15T21:11:50Z","abstract_excerpt":"We present Universal Manipulation Interface (UMI) -- a data collection and policy learning framework that allows direct skill transfer from in-the-wild human demonstrations to deployable robot policies. UMI employs hand-held grippers coupled with careful interface design to enable portable, low-cost, and information-rich data collection for challenging bimanual and dynamic manipulation demonstrations. To facilitate deployable policy learning, UMI incorporates a carefully designed policy interface with inference-time latency matching and a relative-trajectory action representation. The resultin"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"policies learned via UMI zero-shot generalize to novel environments and objects when trained on diverse human demonstrations. The resulting learned policies are hardware-agnostic and deployable across multiple robot platforms.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The carefully designed gripper interface, inference-time latency matching, and relative-trajectory action representation sufficiently minimize the domain gap between human demonstrations and robot execution to enable reliable zero-shot transfer.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"UMI enables zero-shot deployment of robot manipulation policies trained solely on portable human demonstrations captured with custom handheld grippers, supporting dynamic bimanual tasks across novel environments and objects.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"UMI lets robots learn complex manipulation from portable human gripper demonstrations with zero-shot transfer to new settings and hardware.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"bfbea988bf24b41ee51ea2714cac893ea9f5b9a99dcbc828b25c49cd402dd16a"},"source":{"id":"2402.10329","kind":"arxiv","version":3},"verdict":{"id":"8144b9da-f6c2-4f56-9eda-bf9c8737afdf","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T21:31:10.296068Z","strongest_claim":"policies learned via UMI zero-shot generalize to novel environments and objects when trained on diverse human demonstrations. The resulting learned policies are hardware-agnostic and deployable across multiple robot platforms.","one_line_summary":"UMI enables zero-shot deployment of robot manipulation policies trained solely on portable human demonstrations captured with custom handheld grippers, supporting dynamic bimanual tasks across novel environments and objects.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The carefully designed gripper interface, inference-time latency matching, and relative-trajectory action representation sufficiently minimize the domain gap between human demonstrations and robot execution to enable reliable zero-shot transfer.","pith_extraction_headline":"UMI lets robots learn complex manipulation from portable human gripper demonstrations with zero-shot transfer to new settings and hardware."},"references":{"count":69,"sample":[{"doi":"","year":2022,"title":"Human-to-robot imitation in the wild","work_id":"193d274f-d3df-4318-a002-a6d7cbd4fa9b","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Affordances from human videos as a versatile representation for robotics","work_id":"b304b1d8-62ac-4d6f-b661-13b44adf5b8c","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Rt-1: Robotics transformer for real-world control at scale","work_id":"5f44a925-acb8-445b-ae0f-217b040e39ef","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2014,"title":"Humanoid robot teleoperation with vibrotactile based balancing feedback","work_id":"27d865ab-2d6f-43de-95a9-d7537df0a486","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1109/icar.2015.7251504","year":2015,"title":"Berk Calli, Arjun Singh, Aaron Walsman, Siddhartha Srinivasa, Pieter Abbeel, and Aaron M. Dollar. The ycb object and model set: Towards common benchmarks for manipulation research. In 2015 Internation","work_id":"4f6d6565-f9b4-4be9-b051-cc3c3e4a83a7","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":69,"snapshot_sha256":"0b7bbe18673cbe98a639e84d87cfa1681798db9651c4a6b4445b49c19cfa536e","internal_anchors":2},"formal_canon":{"evidence_count":3,"snapshot_sha256":"37e18ed4fd1a4d89d4b714d43e120a8ea0719773712e84304a8f661f10fb4a68"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}