{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,1,23]],"date-time":"2026-01-23T22:54:18Z","timestamp":1769208858328,"version":"3.49.0"},"reference-count":79,"publisher":"SAGE Publications","issue":"2-3","license":[{"start":{"date-parts":[[2021,10,3]],"date-time":"2021-10-03T00:00:00Z","timestamp":1633219200000},"content-version":"vor","delay-in-days":365,"URL":"http:\/\/www.sagepub.com\/licence-information-for-chorus"}],"funder":[{"DOI":"10.13039\/100000006","name":"Office of Naval Research","doi-asserted-by":"publisher","award":["N00014- 18-1-2873"],"award-info":[{"award-number":["N00014- 18-1-2873"]}],"id":[{"id":"10.13039\/100000006","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/100000001","name":"National Science Foundation","doi-asserted-by":"publisher","award":["IIS-1755038"],"award-info":[{"award-number":["IIS-1755038"]}],"id":[{"id":"10.13039\/100000001","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/100006785","name":"google","doi-asserted-by":"publisher","id":[{"id":"10.13039\/100006785","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/100008536","name":"amazon web services","doi-asserted-by":"publisher","id":[{"id":"10.13039\/100008536","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":["journals.sagepub.com"],"crossmark-restriction":true},"short-container-title":["The International Journal of Robotics Research"],"published-print":{"date-parts":[[2021,2]]},"abstract":"<jats:p> Our goal is to learn control policies for robots that provably generalize well to novel environments given a dataset of example environments. The key technical idea behind our approach is to leverage tools from generalization theory in machine learning by exploiting a precise analogy (which we present in the form of a reduction) between generalization of control policies to novel environments and generalization of hypotheses in the supervised learning setting. In particular, we utilize the probably approximately correct (PAC)-Bayes framework, which allows us to obtain upper bounds that hold with high probability on the expected cost of (stochastic) control policies across novel environments. We propose policy learning algorithms that explicitly seek to minimize this upper bound. The corresponding optimization problem can be solved using convex optimization (relative entropy programming in particular) in the setting where we are optimizing over a finite policy space. In the more general setting of continuously parameterized policies (e.g., neural network policies), we minimize this upper bound using stochastic gradient descent. We present simulated results of our approach applied to learning (1) reactive obstacle avoidance policies and (2) neural network-based grasping policies. We also present hardware results for the Parrot Swing drone navigating through different obstacle environments. Our examples demonstrate the potential of our approach to provide strong generalization guarantees for robotic systems with continuous state and action spaces, complicated (e.g., nonlinear) dynamics, rich sensory inputs (e.g., depth images), and neural network-based policies. <\/jats:p>","DOI":"10.1177\/0278364920959444","type":"journal-article","created":{"date-parts":[[2020,10,3]],"date-time":"2020-10-03T07:10:53Z","timestamp":1601709053000},"page":"574-593","update-policy":"https:\/\/doi.org\/10.1177\/sage-journals-update-policy","source":"Crossref","is-referenced-by-count":19,"title":["PAC-Bayes control: learning policies that provably generalize to novel environments"],"prefix":"10.1177","volume":"40","author":[{"given":"Anirudha","family":"Majumdar","sequence":"first","affiliation":[{"name":"Department of Mechanical and Aerospace Engineering,Princeton University, Princeton, NJ, USA"}]},{"given":"Alec","family":"Farid","sequence":"additional","affiliation":[{"name":"Department of Mechanical and Aerospace Engineering,Princeton University, Princeton, NJ, USA"}]},{"given":"Anoopkumar","family":"Sonar","sequence":"additional","affiliation":[{"name":"Department of Computer Science, Princeton University,Princeton, NJ, USA"}]}],"member":"179","published-online":{"date-parts":[[2020,10,3]]},"reference":[{"key":"bibr1-0278364920959444","first-page":"5074","author":"Agrawal P","year":"2016","journal-title":"Advances in Neural Information Processing Systems"},{"key":"bibr2-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1109\/IROS.2015.7353861"},{"key":"bibr3-0278364920959444","volume-title":"Behavior-Based Robotics","author":"Arkin RC","year":"1998"},{"key":"bibr4-0278364920959444","author":"Arora S","year":"2018","journal-title":"arXiv preprint arXiv:1802.06509"},{"key":"bibr5-0278364920959444","volume-title":"Learning Decisions: Robustness, Uncertainty, and Approximation","author":"Bagnell JA","year":"2004"},{"key":"bibr6-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1109\/ROBOT.2001.932842"},{"key":"bibr7-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1007\/s10514-009-9139-6"},{"key":"bibr8-0278364920959444","first-page":"7694","author":"Bjorck N","year":"2018","journal-title":"Advances in Neural Information Processing Systems"},{"key":"bibr9-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1109\/ACC.2006.1656653"},{"key":"bibr10-0278364920959444","first-page":"499","volume":"2","author":"Bousquet O","year":"2002","journal-title":"Journal of Machine Learning Research"},{"key":"bibr11-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1017\/CBO9780511804441"},{"key":"bibr12-0278364920959444","first-page":"213","volume":"3","author":"Brafman RI","year":"2002","journal-title":"Journal of Machine Learning Research"},{"key":"bibr13-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1007\/s10107-016-0998-2"},{"key":"bibr14-0278364920959444","volume-title":"ShapeNet: An information-rich 3D model repository","author":"Chang AX","year":"2015"},{"key":"bibr15-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1287\/mnsc.6.1.73"},{"key":"bibr16-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1007\/s10514-009-9140-0"},{"key":"bibr17-0278364920959444","unstructured":"Coumans E, Bai Y (2018) PyBullet, a Python module for physics simulation for games, robotics and machine learning. Available at: https:\/\/pybullet.org\/wordpress\/"},{"key":"bibr18-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2009.5206848"},{"issue":"83","key":"bibr19-0278364920959444","first-page":"1","volume":"17","author":"Diamond S","year":"2016","journal-title":"Journal of Machine Learning Research"},{"key":"bibr20-0278364920959444","doi-asserted-by":"publisher","DOI":"10.23919\/ECC.2013.6669541"},{"key":"bibr21-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1002\/cpa.3160280102"},{"key":"bibr22-0278364920959444","author":"Dziugaite GK","year":"2017","journal-title":"arXiv preprint arXiv:1703.11008"},{"key":"bibr23-0278364920959444","author":"Dziugaite GK","year":"2017","journal-title":"arXiv preprint arXiv:1712.09376"},{"key":"bibr24-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1109\/IROS.2015.7353446"},{"key":"bibr25-0278364920959444","first-page":"1624","author":"Fard MM","year":"2010","journal-title":"Advances in Neural Information Processing Systems"},{"key":"bibr26-0278364920959444","author":"Fard MM","year":"2012","journal-title":"arXiv preprint arXiv:1202.3717"},{"key":"bibr27-0278364920959444","author":"Finn C","year":"2017","journal-title":"arXiv preprint arXiv:1703.03400"},{"key":"bibr28-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1007\/BF00993593"},{"key":"bibr29-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1109\/ROBOT.2007.363138"},{"key":"bibr30-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1007\/BFb0007371"},{"key":"bibr31-0278364920959444","author":"Fu J","year":"2014","journal-title":"arXiv preprint arXiv:1404.7073"},{"key":"bibr32-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1145\/1553374.1553419"},{"key":"bibr33-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1007\/978-1-4419-7970-4"},{"key":"bibr34-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2017.769"},{"key":"bibr35-0278364920959444","author":"Gupta S","year":"2017","journal-title":"arXiv preprint arXiv:1712.08125"},{"key":"bibr36-0278364920959444","author":"Hardt M","year":"2015","journal-title":"arXiv preprint arXiv:1509.01240"},{"key":"bibr37-0278364920959444","author":"Ioffe S","year":"2015","journal-title":"arXiv preprint arXiv:1502.03167"},{"key":"bibr38-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1016\/S0004-3702(98)00023-X"},{"key":"bibr39-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1162\/089976699300016304"},{"key":"bibr40-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1023\/A:1017984413808"},{"key":"bibr41-0278364920959444","first-page":"1001","author":"Kearns MJ","year":"2000","journal-title":"Advances in Neural Information Processing Systems"},{"key":"bibr42-0278364920959444","first-page":"769","author":"Lacasse A","year":"2007","journal-title":"Advances in Neural Information Processing Systems"},{"key":"bibr43-0278364920959444","first-page":"273","volume":"6","author":"Langford J","year":"2005","journal-title":"Journal of Machine Learning Research"},{"key":"bibr44-0278364920959444","first-page":"809","author":"Langford J","year":"2002","journal-title":"Advances in Neural Information Processing Systems"},{"key":"bibr45-0278364920959444","first-page":"439","author":"Langford J","year":"2003","journal-title":"Advances in Neural Information Processing Systems"},{"key":"bibr46-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1177\/0278364914549607"},{"issue":"1","key":"bibr47-0278364920959444","first-page":"1334","volume":"17","author":"Levine S","year":"2016","journal-title":"The Journal of Machine Learning Research"},{"key":"bibr48-0278364920959444","first-page":"8157","author":"Li Y","year":"2018","journal-title":"Advances in Neural Information Processing Systems"},{"key":"bibr49-0278364920959444","author":"Mahler J","year":"2017","journal-title":"arXiv preprint arXiv:1703.09312"},{"key":"bibr50-0278364920959444","volume-title":"Proceedings of the Conference on Robot Learning (CoRL)","author":"Majumdar A","year":"2018"},{"key":"bibr51-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1177\/0278364917712421"},{"key":"bibr52-0278364920959444","author":"Maurer A","year":"2004","journal-title":"arXiv preprint cs\/0411099"},{"key":"bibr53-0278364920959444","author":"McAllester D","year":"2013","journal-title":"arXiv preprint arXiv:1307.2118"},{"key":"bibr54-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1023\/A:1007618624809"},{"key":"bibr55-0278364920959444","unstructured":"McGovern A (2019) PyParrot 1.5.21. https:\/\/github.com\/amymcgovern\/pyparrot."},{"key":"bibr56-0278364920959444","unstructured":"MOSEK ApS (2019) Mosek fusion API for Python 9.0.84(beta). https:\/\/docs.mosek.com\/9.0\/pythonfusion\/index.html."},{"key":"bibr57-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1137\/1.9781611970791"},{"key":"bibr58-0278364920959444","author":"Neu G","year":"2017","journal-title":"arXiv preprint arXiv:1705.07798"},{"key":"bibr59-0278364920959444","first-page":"5949","author":"Neyshabur B","year":"2017","journal-title":"Advances in Neural Information Processing Systems"},{"key":"bibr60-0278364920959444","author":"Neyshabur B","year":"2017","journal-title":"preprint arXiv:1707.09564"},{"key":"bibr61-0278364920959444","author":"Neyshabur B","year":"2014","journal-title":"arXiv preprint arXiv:1412.6614"},{"key":"bibr62-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1109\/TIT.2010.2068870"},{"key":"bibr63-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1007\/s10957-016-0892-3"},{"key":"bibr64-0278364920959444","unstructured":"O\u2019Donoghue B, Chu E, Parikh N, Boyd S (2017) SCS: Splitting conic solver, version 2.0.2. https:\/\/github.com\/cvxgrp\/scs."},{"key":"bibr65-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1007\/s10514-015-9467-7"},{"key":"bibr66-0278364920959444","doi-asserted-by":"publisher","DOI":"10.15607\/RSS.2017.XIII.064"},{"key":"bibr67-0278364920959444","volume-title":"Proceedings of the International Symposium on Robotics Research (ISRR)","author":"Richter C","year":"2015"},{"key":"bibr68-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1109\/ICRA.2013.6630809"},{"key":"bibr69-0278364920959444","doi-asserted-by":"publisher","DOI":"10.23919\/ACC.2004.1384742"},{"key":"bibr70-0278364920959444","first-page":"1889","volume-title":"Proceedings of the International Conference on Machine Learning","author":"Schulman J","year":"2015"},{"key":"bibr71-0278364920959444","doi-asserted-by":"crossref","first-page":"233","DOI":"10.1162\/153244303765208377","volume":"3","author":"Seeger M","year":"2002","journal-title":"Journal of Machine Learning Research"},{"key":"bibr72-0278364920959444","author":"Simonyan K","year":"2014","journal-title":"CoRR"},{"issue":"1","key":"bibr73-0278364920959444","first-page":"1929","volume":"15","author":"Srivastava N","year":"2014","journal-title":"The Journal of Machine Learning Research"},{"key":"bibr74-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1177\/0278364918770733"},{"key":"bibr75-0278364920959444","author":"Tobin J","year":"2017","journal-title":"arXiv preprint arXiv:1710.06425"},{"key":"bibr76-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1109\/ICRA.2011.5980257"},{"key":"bibr77-0278364920959444","unstructured":"Wolfram, Research Inc. (2019) Mathematica, Version 12.0. https:\/\/www.wolfram.com\/mathematica\/."},{"key":"bibr78-0278364920959444","author":"Zhang C","year":"2016","journal-title":"arXiv preprint arXiv:1611.03530"},{"key":"bibr79-0278364920959444","doi-asserted-by":"publisher","DOI":"10.1109\/ICRA.2017.7989381"}],"container-title":["The International Journal of Robotics Research"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/journals.sagepub.com\/doi\/pdf\/10.1177\/0278364920959444","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/journals.sagepub.com\/doi\/full-xml\/10.1177\/0278364920959444","content-type":"application\/xml","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/journals.sagepub.com\/doi\/pdf\/10.1177\/0278364920959444","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"https:\/\/journals.sagepub.com\/doi\/pdf\/10.1177\/0278364920959444","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,3,1]],"date-time":"2025-03-01T15:17:58Z","timestamp":1740842278000},"score":1,"resource":{"primary":{"URL":"https:\/\/journals.sagepub.com\/doi\/10.1177\/0278364920959444"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2020,10,3]]},"references-count":79,"journal-issue":{"issue":"2-3","published-print":{"date-parts":[[2021,2]]}},"alternative-id":["10.1177\/0278364920959444"],"URL":"https:\/\/doi.org\/10.1177\/0278364920959444","relation":{},"ISSN":["0278-3649","1741-3176"],"issn-type":[{"value":"0278-3649","type":"print"},{"value":"1741-3176","type":"electronic"}],"subject":[],"published":{"date-parts":[[2020,10,3]]}}}