{"paper":{"title":"MimicGen: A Data Generation System for Scalable Robot Learning using Human Demonstrations","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"MimicGen adapts a few hundred human demonstrations into over 50,000 varied examples that train robots for long-horizon tasks.","cross_cats":["cs.AI","cs.CV","cs.LG"],"primary_cat":"cs.RO","authors_text":"Ajay Mandlekar, Bowen Wen, Dieter Fox, Iretiayo Akinola, Linxi Fan, Soroush Nasiriany, Yashraj Narang, Yuke Zhu","submitted_at":"2023-10-26T17:17:31Z","abstract_excerpt":"Imitation learning from a large set of human demonstrations has proved to be an effective paradigm for building capable robot agents. However, the demonstrations can be extremely costly and time-consuming to collect. We introduce MimicGen, a system for automatically synthesizing large-scale, rich datasets from only a small number of human demonstrations by adapting them to new contexts. We use MimicGen to generate over 50K demonstrations across 18 tasks with diverse scene configurations, object instances, and robot arms from just ~200 human demonstrations. We show that robot agents can be effe"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We show that robot agents can be effectively trained on this generated dataset by imitation learning to achieve strong performance in long-horizon and high-precision tasks, such as multi-part assembly and coffee preparation, across broad initial state distributions.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"Adapting human demonstrations to new contexts produces data that is as effective for training as real human demonstrations collected in those contexts.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"MimicGen creates over 50K robot demonstrations from roughly 200 human ones, allowing imitation learning to achieve strong performance on complex long-horizon tasks like assembly and coffee preparation.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"MimicGen adapts a few hundred human demonstrations into over 50,000 varied examples that train robots for long-horizon tasks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"d86744b0cfe07160d252e47089cfcc719f9e5b88e63e78c584a7a8c6c39c6ab8"},"source":{"id":"2310.17596","kind":"arxiv","version":1},"verdict":{"id":"c47a1c43-0807-4dcc-bfa0-9d2598d7cae5","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T09:41:01.720309Z","strongest_claim":"We show that robot agents can be effectively trained on this generated dataset by imitation learning to achieve strong performance in long-horizon and high-precision tasks, such as multi-part assembly and coffee preparation, across broad initial state distributions.","one_line_summary":"MimicGen creates over 50K robot demonstrations from roughly 200 human ones, allowing imitation learning to achieve strong performance on complex long-horizon tasks like assembly and coffee preparation.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"Adapting human demonstrations to new contexts produces data that is as effective for training as real human demonstrations collected in those contexts.","pith_extraction_headline":"MimicGen adapts a few hundred human demonstrations into over 50,000 varied examples that train robots for long-horizon tasks."},"references":{"count":128,"sample":[{"doi":"","year":2017,"title":"Deep Imitation Learning for Complex Manipulation Tasks from Virtual Reality Teleoperation","work_id":"09d7abfe-a94e-47ef-8064-2329faccb197","ref_index":1,"cited_arxiv_id":"1710.04615","is_internal_anchor":true},{"doi":"","year":2018,"title":"RoboTurk: A Crowdsourcing Platform for Robotic Skill Learning through Imitation,","work_id":"1f0e1f61-6910-4924-8f48-93cc372a3dc2","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Bc- z: Zero-shot task generalization with robotic imitation learning,","work_id":"fd91f703-9aa1-495a-951b-403bced21c97","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Do As I Can, Not As I Say: Grounding Language in Robotic Affordances","work_id":"037320f1-b0a9-4cbe-a639-bfb25409ce71","ref_index":4,"cited_arxiv_id":"2204.01691","is_internal_anchor":true},{"doi":"","year":2022,"title":"RT-1: Robotics Transformer for Real-World Control at Scale","work_id":"e11bda85-8531-46bc-a07f-d0ade3643ab1","ref_index":5,"cited_arxiv_id":"2212.06817","is_internal_anchor":true}],"resolved_work":128,"snapshot_sha256":"7ccca6e98ac266610acdb424a6d4be467ef44a4d81890659e227381e511e340a","internal_anchors":13},"formal_canon":{"evidence_count":2,"snapshot_sha256":"91573c6ce45c5cc0765f91da94fed9634d7cee69b60a404aa32c65824f58d38e"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}