Task-Driven Detection of Distribution Shifts With Statistical Guarantees for Robot Learning

Our goal is to perform out-of-distribution (OOD) detection , i.e., to detect when a robot is operating in environments drawn from a different distribution than the ones used to train the robot. We leverage probably approximately correct-Bayes theory to train a policy with a guaranteed bound on perfo...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:IEEE transactions on robotics Jg. 41; S. 926 - 945
Hauptverfasser: Farid, Alec, Veer, Sushant, Pachisia, Divyanshu, Majumdar, Anirudha
Format: Journal Article
Sprache:Englisch
Veröffentlicht: IEEE 2025
Schlagworte:
ISSN:1552-3098, 1941-0468
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Abstract Our goal is to perform out-of-distribution (OOD) detection , i.e., to detect when a robot is operating in environments drawn from a different distribution than the ones used to train the robot. We leverage probably approximately correct-Bayes theory to train a policy with a guaranteed bound on performance on the training distribution. Our idea for OOD detection relies on the following intuition: violation of the performance bound on test environments provides evidence that the robot is operating OOD. We formalize this via statistical techniques based on <inline-formula><tex-math notation="LaTeX">p</tex-math></inline-formula>-values and concentration inequalities. The approach provides guaranteed confidence bounds on OOD detection including bounds on both the false-positive and false-negative rates of the detector and is task-driven and only sensitive to changes that impact the robot's performance. We demonstrate our approach in simulation and hardware for a grasping task using objects with unfamiliar shapes or poses and a drone performing vision-based obstacle avoidance in environments with wind disturbances and varied obstacle densities. Our examples demonstrate that we can perform task-driven OOD detection within just a handful of trials.
AbstractList Our goal is to perform out-of-distribution (OOD) detection , i.e., to detect when a robot is operating in environments drawn from a different distribution than the ones used to train the robot. We leverage probably approximately correct-Bayes theory to train a policy with a guaranteed bound on performance on the training distribution. Our idea for OOD detection relies on the following intuition: violation of the performance bound on test environments provides evidence that the robot is operating OOD. We formalize this via statistical techniques based on <inline-formula><tex-math notation="LaTeX">p</tex-math></inline-formula>-values and concentration inequalities. The approach provides guaranteed confidence bounds on OOD detection including bounds on both the false-positive and false-negative rates of the detector and is task-driven and only sensitive to changes that impact the robot's performance. We demonstrate our approach in simulation and hardware for a grasping task using objects with unfamiliar shapes or poses and a drone performing vision-based obstacle avoidance in environments with wind disturbances and varied obstacle densities. Our examples demonstrate that we can perform task-driven OOD detection within just a handful of trials.
Author Pachisia, Divyanshu
Majumdar, Anirudha
Veer, Sushant
Farid, Alec
Author_xml – sequence: 1
  givenname: Alec
  orcidid: 0000-0003-4253-496X
  surname: Farid
  fullname: Farid, Alec
  email: afarid@zoox.com
  organization: Princeton University, Princeton, NJ, USA
– sequence: 2
  givenname: Sushant
  orcidid: 0000-0003-3498-6448
  surname: Veer
  fullname: Veer, Sushant
  email: sveer@nvidia.com
  organization: Princeton University, Princeton, NJ, USA
– sequence: 3
  givenname: Divyanshu
  orcidid: 0009-0007-5120-1899
  surname: Pachisia
  fullname: Pachisia, Divyanshu
  email: divyanshu.pachisia@skydio.com
  organization: Princeton University, Princeton, NJ, USA
– sequence: 4
  givenname: Anirudha
  orcidid: 0009-0002-2296-7485
  surname: Majumdar
  fullname: Majumdar, Anirudha
  email: ani.majumdar@princeton.edu
  organization: Intelligent Robot Motion Lab, Princeton University, Princeton, NJ, USA
BookMark eNpNkE9LAzEQxYNUsK3ePXjIF9iaf5tsjtJqFQqFtuLBw5LdTGy0biRJBb-9W9uDh2FmmPcew2-EBl3oAKFrSiaUEn27WS0njDAx4SWjWvIzNKRa0IIIWQ36uSxZwYmuLtAopXfSKzXhQ_S6MemjmEX_DR2eQYY2-9Dh4PDMpxx9s__b11vvcsIvPm_xOpvc33xrdni-N9F0GSBhFyJehSZkvAATO9-9XaJzZ3YJrk59jJ4f7jfTx2KxnD9N7xZFy6jKhTT9M1Iwx4ALkMRSVQErdaWakjhjrQaltHKi0YYKJ0trrRFKG6O0VJbzMSLH3DaGlCK4-iv6TxN_akrqA5y6h1Mf4NQnOL3l5mjxAPBPXtGyL_4LAUBjPQ
CODEN ITREAE
Cites_doi 10.1017/CBO9781107298019
10.1109/JPROC.2021.3052449
10.1201/9781003140351
10.15607/RSS.2022.XVIII.042
10.1145/267460.267466
10.1145/1553374.1553419
10.1214/22-AOS2244
10.15607/rss.2017.xiii.064
10.1023/A:1007618624809
10.1016/j.ins.2013.06.018
10.1007/978-3-030-61616-8_34
10.1109/ICCPS48487.2020.00024
10.1177/0278364920959444
10.1109/CVPR42600.2020.01401
10.1007/b99352
10.1016/0005-1098(88)90073-8
ContentType Journal Article
DBID 97E
RIA
RIE
AAYXX
CITATION
DOI 10.1109/TRO.2024.3521963
DatabaseName IEEE Xplore (IEEE)
IEEE All-Society Periodicals Package (ASPP) 1998–Present
IEEE Electronic Library (IEL)
CrossRef
DatabaseTitle CrossRef
DatabaseTitleList
Database_xml – sequence: 1
  dbid: RIE
  name: IEEE Electronic Library (IEL)
  url: https://ieeexplore.ieee.org/
  sourceTypes: Publisher
DeliveryMethod fulltext_linktorsrc
Discipline Engineering
EISSN 1941-0468
EndPage 945
ExternalDocumentID 10_1109_TRO_2024_3521963
10815081
Genre orig-research
GrantInformation_xml – fundername: Toyota Research Institute (TRI)
– fundername: NSF CAREER Award
  grantid: 2044149
– fundername: Office of Naval Research
  grantid: N00014-21-1-2803; N00014-18-1-2873
  funderid: 10.13039/100000006
GroupedDBID .DC
0R~
29I
4.4
5GY
5VS
6IK
97E
AAJGR
AARMG
AASAJ
AAWTH
ABAZT
ABQJQ
ABVLG
ACGFO
ACIWK
AENEX
AETIX
AGQYO
AGSQL
AHBIQ
AIBXA
AKJIK
AKQYR
ALMA_UNASSIGNED_HOLDINGS
ATWAV
BEFXN
BFFAM
BGNUA
BKEBE
BPEOZ
CS3
DU5
EBS
EJD
F5P
HZ~
H~9
IFIPE
IPLJI
JAVBF
LAI
M43
MS~
O9-
OCL
P2P
PQQKQ
RIA
RIE
RNS
VJK
AAYXX
CITATION
ID FETCH-LOGICAL-c217t-6a024642f2e34e60d178e25987b50fadd9e7797f4b9a14f65ddda479aa7967d33
IEDL.DBID RIE
ISICitedReferencesCount 0
ISICitedReferencesURI http://www.webofscience.com/api/gateway?GWVersion=2&SrcApp=Summon&SrcAuth=ProQuest&DestLinkType=CitingArticles&DestApp=WOS_CPL&KeyUT=001398617800002&url=https%3A%2F%2Fcvtisr.summon.serialssolutions.com%2F%23%21%2Fsearch%3Fho%3Df%26include.ft.matches%3Dt%26l%3Dnull%26q%3D
ISSN 1552-3098
IngestDate Sat Nov 29 01:47:35 EST 2025
Wed Aug 27 01:55:50 EDT 2025
IsPeerReviewed true
IsScholarly true
Language English
License https://ieeexplore.ieee.org/Xplorehelp/downloads/license-information/IEEE.html
https://doi.org/10.15223/policy-029
https://doi.org/10.15223/policy-037
LinkModel DirectLink
MergedId FETCHMERGED-LOGICAL-c217t-6a024642f2e34e60d178e25987b50fadd9e7797f4b9a14f65ddda479aa7967d33
ORCID 0009-0007-5120-1899
0009-0002-2296-7485
0000-0003-3498-6448
0000-0003-4253-496X
PageCount 20
ParticipantIDs crossref_primary_10_1109_TRO_2024_3521963
ieee_primary_10815081
PublicationCentury 2000
PublicationDate 20250000
2025-00-00
PublicationDateYYYYMMDD 2025-01-01
PublicationDate_xml – year: 2025
  text: 20250000
PublicationDecade 2020
PublicationTitle IEEE transactions on robotics
PublicationTitleAbbrev TRO
PublicationYear 2025
Publisher IEEE
Publisher_xml – name: IEEE
References Bitterwolf (ref13) 2020
ref56
Catoni (ref40) 2007; 56
Lee (ref12) 2018
Meinke (ref15) 2021
ref10
ref54
ref19
Jiang (ref36) 2020
Ren (ref47) 2020
Thiemann (ref43) 2017; 76
ref48
Greenberg (ref26) 2021
Meinke (ref14) 2020
Veer (ref46) 2020
Langford (ref33) 2003
Prez-Ortiz (ref37) 2021; 22
ref3
Vapnik (ref27) 1968; 181
Fang (ref18) 2024; 25
Hendrycks (ref9) 2019
Farid (ref1) 2021
ref34
ref30
Maurer (ref55) 2004
ref2
ref39
Dziugaite (ref32) 2017
Lee (ref8) 2018
DeVries (ref7) 2018
Sedlmeier (ref22) 2019
Chen (ref11) 2020
Hogg (ref51) 2005
Hendrycks (ref6) 2022
ref24
ref23
Liang (ref5) 2018
Bartlett (ref35) 2017
ref20
ref21
Seeger (ref31) 2002; 3
Rivasplata (ref42) 2019
Coumans (ref52) 2018
ref28
Lotfi (ref38) 2022; 35
Dziugaite (ref44) 2018
Majumdar (ref45) 2018; 87
ref29
Viallard (ref49) 2021
Chang (ref53) 2015
McAllester (ref41) 2013
Liu (ref17) 2018
Hendrycks (ref4) 2017
Wu (ref25) 2021
Siddiqui (ref16) 2016
Wierstra (ref50) 2014; 15
References_xml – ident: ref28
  doi: 10.1017/CBO9781107298019
– year: 2015
  ident: ref53
  article-title: Shapenet: An information-rich 3D model repository
– volume-title: Proc. Int. Conf. Learn. Representations
  year: 2019
  ident: ref9
  article-title: Deep anomaly detection with outlier exposure
– ident: ref3
  doi: 10.1109/JPROC.2021.3052449
– volume-title: Proc. Int. Conf. Learn. Representations
  year: 2018
  ident: ref5
  article-title: Enhancing the reliability of out-of-distribution image detection in neural networks
– start-page: 970
  volume-title: Proc. Conf. Robot Learn.
  year: 2021
  ident: ref1
  article-title: Task-driven out-of-distribution detection with statistical guarantees for robot learning
– volume-title: Proc. Int. Conf. Learn. Representations
  year: 2017
  ident: ref4
  article-title: A baseline for detecting misclassified and out-of-distribution examples in neural networks
– ident: ref20
  doi: 10.1201/9781003140351
– ident: ref54
  doi: 10.15607/RSS.2022.XVIII.042
– volume-title: Proc. Int. Conf. Learn. Representations
  year: 2020
  ident: ref36
  article-title: Fantastic generalization measures and where to find them
– ident: ref29
  doi: 10.1145/267460.267466
– start-page: 439
  volume-title: Proc. Conf. Adv. Neural Inf. Process. Syst.
  year: 2003
  ident: ref33
  article-title: PAC-Bayes & margins
– ident: ref34
  doi: 10.1145/1553374.1553419
– volume: 76
  start-page: 1
  year: 2017
  ident: ref43
  article-title: A strongly quasiconvex PAC-Bayesian bound
  publication-title: Mach. Learn. Res.
– volume-title: Proc. Conf. Adv. Neural Inf. Process. Syst.
  year: 2018
  ident: ref44
  article-title: Data-dependent PAC-Bayes priors via differential privacy
– year: 2018
  ident: ref7
  article-title: Learning confidence for out-of-distribution detection in neural networks
– year: 2021
  ident: ref15
  article-title: Provably robust detection of out-of-distribution data (almost) for free
– year: 2018
  ident: ref52
  article-title: Pybullet, a python module for physics simulation for games, robotics and machine learning
– year: 2004
  ident: ref55
  article-title: A note on the PAC Bayesian theorem
– ident: ref19
  doi: 10.1214/22-AOS2244
– volume: 3
  start-page: 233
  year: 2002
  ident: ref31
  article-title: PAC-Bayesian generalisation error bounds for Gaussian process classification
  publication-title: J. Mach. Learn. Res.
– start-page: 16085
  volume-title: Proc. Conf. Adv. Neural Inf. Process. Syst.
  year: 2020
  ident: ref13
  article-title: Certifiably adversarially robust detection of out-of-distribution data
– start-page: 8759
  volume-title: Proc. Int. Conf. Mach. Learn.
  year: 2022
  ident: ref6
  article-title: Scaling out-of-distribution detection for real-world settings
– volume: 35
  start-page: 31459
  volume-title: Proc. Int. Conf. Adv. Neural Inf. Process. Syst.
  year: 2022
  ident: ref38
  article-title: PAC-Bayes compression bounds so tight that they can explain generalization
– volume-title: Introduction to Mathematical Statistics
  year: 2005
  ident: ref51
– ident: ref21
  doi: 10.15607/rss.2017.xiii.064
– start-page: 1001
  volume-title: Proc. Conf. Robot Learn.
  year: 2020
  ident: ref46
  article-title: Probably approximately correct vision-based planning using motion primitives
– ident: ref30
  doi: 10.1023/A:1007618624809
– volume: 22
  start-page: 1
  year: 2021
  ident: ref37
  article-title: Tighter risk certificates for neural networks
  publication-title: J. Mach. Learn. Res.
– ident: ref56
  doi: 10.1016/j.ins.2013.06.018
– year: 2013
  ident: ref41
  article-title: A PAC-Bayesian tutorial with a dropout bound
– ident: ref23
  doi: 10.1007/978-3-030-61616-8_34
– start-page: 1426
  volume-title: Proc. Conf. Robot Learn.
  year: 2020
  ident: ref47
  article-title: Generalization guarantees for imitation learning
– year: 2019
  ident: ref42
  article-title: PAC-Bayes with backprop
– ident: ref24
  doi: 10.1109/ICCPS48487.2020.00024
– volume-title: Proc. Int. Conf. Learn. Representations
  year: 2018
  ident: ref8
  article-title: Training confidence-calibrated classifiers for detecting out-of-distribution samples
– start-page: 686
  volume-title: Proc. Conf. Uncertainty Artif. Intell.
  year: 2016
  ident: ref16
  article-title: Finite sample complexity of rare pattern anomaly detection
– volume: 181
  start-page: 781
  issue: 4
  year: 1968
  ident: ref27
  article-title: On the uniform convergence of relative frequencies of events to their probabilities
  publication-title: Doklady Akademii Nauk USSR
– start-page: 7167
  volume-title: Proc. Conf. Adv. Neural Inf. Process. Syst.
  year: 2018
  ident: ref12
  article-title: A simple unified framework for detecting out-of-distribution samples and adversarial attacks
– volume-title: Proc. Int. Conf. Learn. Representations
  year: 2020
  ident: ref14
  article-title: Towards neural networks that provably know when they don’t know
– volume: 87
  start-page: 293
  volume-title: Proc. Conf. Robot Learn.
  year: 2018
  ident: ref45
  article-title: PAC-Bayes control: Synthesizing controllers that provably generalize to novel environments
– year: 2019
  ident: ref22
  article-title: Uncertainty-based out-of-distribution classification in deep reinforcement learning
– ident: ref48
  doi: 10.1177/0278364920959444
– volume-title: Proc. Conf. Uncertainty Artif. Intell.
  year: 2017
  ident: ref32
  article-title: Computing nonvacuous generalization bounds for deep (Stochastic) neural networks with many more parameters than training data
– start-page: 3169
  volume-title: Proc. Int. Conf. Mach. Learn.
  year: 2018
  ident: ref17
  article-title: Open category detection with PAC guarantees
– volume: 25
  start-page: 1
  year: 2024
  ident: ref18
  article-title: On the learnability of out-of-distribution detection
  publication-title: J. Mach. Learn. Res.
– year: 2021
  ident: ref25
  article-title: RLAD: Time series anomaly detection through reinforcement learning and active learning
– volume: 15
  start-page: 949
  issue: 27
  year: 2014
  ident: ref50
  article-title: Natural evolution strategies
  publication-title: J. Mach. Learn. Res.
– volume: 56
  volume-title: PAC-Bayesian Supervised Classification: The Thermodynamics of Statistical Learning
  year: 2007
  ident: ref40
– ident: ref10
  doi: 10.1109/CVPR42600.2020.01401
– ident: ref39
  doi: 10.1007/b99352
– year: 2021
  ident: ref49
  article-title: A general framework for the derandomization of PAC-Bayesian bounds
– start-page: 6240
  volume-title: Proc. Int. Conf. Adv. Neural Inf. Process. Syst.
  year: 2017
  ident: ref35
  article-title: Spectrally-normalized margin bounds for neural networks
– start-page: 3842
  volume-title: Proc. Int. Conf. Mach. Learn.
  year: 2021
  ident: ref26
  article-title: Detecting rewards deterioration in episodic reinforcement learning
– ident: ref2
  doi: 10.1016/0005-1098(88)90073-8
– year: 2020
  ident: ref11
  article-title: Robust out-of-distribution detection for neural networks
SSID ssj0024903
Score 2.4492779
Snippet Our goal is to perform out-of-distribution (OOD) detection , i.e., to detect when a robot is operating in environments drawn from a different distribution than...
SourceID crossref
ieee
SourceType Index Database
Publisher
StartPage 926
SubjectTerms Collision avoidance
Costs
Deep learning in robotics and automation
Detectors
Drones
failure detection and recovery
formal methods in robotics and automation
Grasping
Hardware
Navigation
PAC-Bayes
Robot sensing systems
Robots
Training
Title Task-Driven Detection of Distribution Shifts With Statistical Guarantees for Robot Learning
URI https://ieeexplore.ieee.org/document/10815081
Volume 41
WOSCitedRecordID wos001398617800002&url=https%3A%2F%2Fcvtisr.summon.serialssolutions.com%2F%23%21%2Fsearch%3Fho%3Df%26include.ft.matches%3Dt%26l%3Dnull%26q%3D
hasFullText 1
inHoldings 1
isFullTextHit
isPrint
journalDatabaseRights – providerCode: PRVIEE
  databaseName: IEEE Electronic Library (IEL)
  customDbUrl:
  eissn: 1941-0468
  dateEnd: 99991231
  omitProxy: false
  ssIdentifier: ssj0024903
  issn: 1552-3098
  databaseCode: RIE
  dateStart: 20040101
  isFulltext: true
  titleUrlDefault: https://ieeexplore.ieee.org/
  providerName: IEEE
link http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV09T8MwELWgYoCBzyLKlzywMLg4jZuzR0RBDKigUkQlhsiNbVohJahN-f2cnVSUgYEtiqIoebn43rN97wi5kJCB4caxsTJdJoQBppQzzGVOJkpjCLlQKPwA_b4cjdRTXaweamGstWHzmW37w7CWb4ps4afK8A-X3r4cxc46AFTFWj_Geiq0QfaWYizmSi7XJLm6Gg4eUQl2RBvZho-4XzlopalKyCl3O_98ml2yXZNHel197T2yZvN9srViKXhA3oZ6_sF6Mz-I0Z4tw06rnBaO9rxFbt3dij5Ppq6c09dpOaGebwa7Zry1jxiPtZ1TZLN0UIyLktYerO9N8nJ3O7y5Z3UDBZah0ihZovGlUWC4jo2FTbiJQFrUOxLGXe5wZFMWQIETY6Uj4ZKuMUYLUFqDSsDE8SFp5EVujwhF1YRcxlkls1joyGlPNADTP08M11HUIpdLSNPPyicjDfqCqxThTz38aQ1_izQ9mivXVUAe_3H-hGx2fNfdMPFxShrlbGHPyEb2heDMzkMUfAMSwLBg
linkProvider IEEE
linkToHtml http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV1NS8MwGA4yBfXg58T5mYMXD5n9SJvmKM4xcU6ZFQceStokbgitbJ2_3zdph_PgwVsJpTRP3ybPk-R9XoQuIpYx6UhNUi4DQqlkhHMtic50FHIBIaRtonCfDQbRaMSf6mR1mwujlLKHz1TbXNq9fFlkc7NUBn94ZOzLQeysBpR6bpWu9WOtx20hZGMqRnyHR4tdSYdfxcNH0IIebQPfMDH3axZaKqtiZ5Xu9j_fZwdt1fQRX1ffexetqHwPbS6ZCu6jt1jMPkhnaoYx3FGlPWuV40LjjjHJretb4efxRJcz_Dopx9gwTmvYDI82MWPQVjMMfBYPi7Qoce3C-t5EL93b-KZH6hIKJAOtUZJQQKdBYmhP-VSFjnRZpEDxRCwNHA1jG1eMcaZpyoVLdRhIKQVlXAjGQyZ9_wA18iJXhwiDbgI2oxWPMp8KVwtDNRgQACeUjnDdFrpcQJp8Vk4ZiVUYDk8A_sTAn9Twt1DToLl0XwXk0R_t52i9Fz_0k_7d4P4YbXimBq9dBjlBjXI6V6doLfsCoKZnNiK-AXADs6c
openUrl ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=Task-Driven+Detection+of+Distribution+Shifts+With+Statistical+Guarantees+for+Robot+Learning&rft.jtitle=IEEE+transactions+on+robotics&rft.au=Farid%2C+Alec&rft.au=Veer%2C+Sushant&rft.au=Pachisia%2C+Divyanshu&rft.au=Majumdar%2C+Anirudha&rft.date=2025&rft.issn=1552-3098&rft.eissn=1941-0468&rft.volume=41&rft.spage=926&rft.epage=945&rft_id=info:doi/10.1109%2FTRO.2024.3521963&rft.externalDBID=n%2Fa&rft.externalDocID=10_1109_TRO_2024_3521963
thumbnail_l http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/lc.gif&issn=1552-3098&client=summon
thumbnail_m http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/mc.gif&issn=1552-3098&client=summon
thumbnail_s http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/sc.gif&issn=1552-3098&client=summon