{"paper":{"title":"RoboNet: Large-Scale Multi-Robot Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Pre-training on a shared dataset from seven robots lets new arms learn tasks with far less data than training from scratch on the target platform alone.","cross_cats":["cs.CV","cs.LG"],"primary_cat":"cs.RO","authors_text":"Bernadette Bucher, Chelsea Finn, Frederik Ebert, Karl Schmeckpeper, Sergey Levine, Siddharth Singh, Stephen Tian, Sudeep Dasari, Suraj Nair","submitted_at":"2019-10-24T15:20:03Z","abstract_excerpt":"Robot learning has emerged as a promising tool for taming the complexity and diversity of the real world. Methods based on high-capacity models, such as deep networks, hold the promise of providing effective generalization to a wide range of open-world environments. However, these same methods typically require large amounts of diverse training data to generalize effectively. In contrast, most robotic learning experiments are small-scale, single-domain, and single-robot. This leads to a frequent tension in robotic learning: how can we learn generalizable robotic controllers without having to c"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"by pre-training on RoboNet and fine-tuning on data from a held-out Franka or Kuka robot, we can exceed the performance of a robot-specific training approach that uses 4x-20x more data.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That visual features and dynamics learned across the seven source robots transfer meaningfully to a held-out robot without large unmodeled domain gaps in gripper mechanics, camera calibration, or task distribution.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"RoboNet is a multi-robot video dataset that enables pre-training of vision-based manipulation models which, after fine-tuning on a new robot, outperform robot-specific training that uses 4-20 times more data.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Pre-training on a shared dataset from seven robots lets new arms learn tasks with far less data than training from scratch on the target platform alone.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"2b0529c102518f2408aa44f6db5887f74b8a8c0557f76559442a94514cd104b3"},"source":{"id":"1910.11215","kind":"arxiv","version":2},"verdict":{"id":"bd90e1f9-90cf-4c51-b190-9d31ab2dad79","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T19:01:53.961231Z","strongest_claim":"by pre-training on RoboNet and fine-tuning on data from a held-out Franka or Kuka robot, we can exceed the performance of a robot-specific training approach that uses 4x-20x more data.","one_line_summary":"RoboNet is a multi-robot video dataset that enables pre-training of vision-based manipulation models which, after fine-tuning on a new robot, outperform robot-specific training that uses 4-20 times more data.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That visual features and dynamics learned across the seven source robots transfer meaningfully to a held-out robot without large unmodeled domain gaps in gripper mechanics, camera calibration, or task distribution.","pith_extraction_headline":"Pre-training on a shared dataset from seven robots lets new arms learn tasks with far less data than training from scratch on the target platform alone."},"references":{"count":60,"sample":[{"doi":"","year":2016,"title":"arXiv preprint arXiv:1611.04201 , year=","work_id":"52f29e55-b23c-42e4-8187-a5ea46200858","ref_index":1,"cited_arxiv_id":"1611.04201","is_internal_anchor":true},{"doi":"","year":2018,"title":"Pachocki, Jakub Pachocki, Arthur Petron, Matthias Plappert, Glenn Powell, Alex Ray, Jonas Schneider, Szymon Sidor, Josh Tobin, Peter Welinder, Lilian Weng, and Wojciech Zaremba","work_id":"389579f0-1ebf-484d-b67d-d8239d049ece","ref_index":2,"cited_arxiv_id":"1808.00177","is_internal_anchor":true},{"doi":"","year":2011,"title":"M. Deisenroth and C. E. Rasmussen. Pilco: A model-based and data-efficient approach to policy search. In International Conference on machine learning (ICML), 2011","work_id":"8b738e5e-bb8f-4b4e-94e1-02e263143af6","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2013,"title":"M. P. Deisenroth, D. Fox, and C. E. Rasmussen. Gaussian processes for data-efficient learning in robotics and control. IEEE transactions on pattern analysis and machine intelligence, 37(2):408–423, 201","work_id":"a26fbba2-ad04-4b9e-b81c-8c5878e9853c","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2016,"title":"C. Finn, I. Goodfellow, and S. Levine. Unsupervised learning for physical interaction through video prediction. In Advances in neural information processing systems, pages 64–72, 2016","work_id":"266915f0-3afa-468c-9b3e-fc2a90d7cbd8","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":60,"snapshot_sha256":"4544f88895bf1791a76861174c333d79e468fdc443821acc3aa0d768bb71c1bb","internal_anchors":25},"formal_canon":{"evidence_count":2,"snapshot_sha256":"801a81ea043e544dc19d5c56e783620f0c7a4c8c0bee5525b47c9957c74e3d71"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}