Temporal-Spatial Causal Interpretations for Vision-Based Reinforcement Learning

被引:7
|
作者
Shi, Wenjie [1 ]
Huang, Gao [1 ,2 ]
Song, Shiji [1 ]
Wu, Cheng [1 ]
机构
[1] Tsinghua Univ, Beijing Natl Res Ctr Informat Sci & Technol BNRis, Dept Automat, Beijing 100084, Peoples R China
[2] Beijing Acad Artificial Intelligence BAAI, Beijing 100084, Peoples R China
基金
中国国家自然科学基金;
关键词
Adaptation models; Reliability; Decision making; Perturbation methods; Visualization; Task analysis; Feature extraction; Reinforcement learning; markov decision process; interpretability; attention map; temporal causality;
D O I
10.1109/TPAMI.2021.3133717
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Deep reinforcement learning (RL) agents are becoming increasingly proficient in a range of complex control tasks. However, the agent's behavior is usually difficult to interpret due to the introduction of black-box function, making it difficult to acquire the trust of users. Although there have been some interesting interpretation methods for vision-based RL, most of them cannot uncover temporal causal information, raising questions about their reliability. To address this problem, we present a temporal-spatial causal interpretation (TSCI) model to understand the agent's long-term behavior, which is essential for sequential decision-making. TSCI model builds on the formulation of temporal causality, which reflects the temporal causal relations between sequential observations and decisions of RL agent. Then a separate causal discovery network is employed to identify temporal-spatial causal features, which are constrained to satisfy the temporal causality. TSCI model is applicable to recurrent agents and can be used to discover causal features with high efficiency once trained. The empirical results show that TSCI model can produce high-resolution and sharp attention masks to highlight task-relevant temporal-spatial information that constitutes most evidence about how vision-based RL agents make sequential decisions. In addition, we further demonstrate that our method is able to provide valuable causal interpretations for vision-based RL agents from the temporal perspective.
引用
收藏
页码:10222 / 10235
页数:14
相关论文
共 50 条
  • [1] Neuromorphic Vision-Based Fall Localization in Event Streams With Temporal-Spatial Attention Weighted Network
    Chen, Guang
    Qu, Sanqing
    Li, Zhijun
    Zhu, Haitao
    Dong, Jiaxuan
    Liu, Min
    Conradt, Joerg
    [J]. IEEE TRANSACTIONS ON CYBERNETICS, 2022, 52 (09) : 9251 - 9262
  • [2] Vision-based reinforcement learning for robot navigation
    Zhu, WY
    Levinson, S
    [J]. IJCNN'01: INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS, VOLS 1-4, PROCEEDINGS, 2001, : 1025 - 1030
  • [3] Vision-based Navigation Using Deep Reinforcement Learning
    Kulhanek, Jonas
    Derner, Erik
    de Bruin, Tim
    Babuska, Robert
    [J]. 2019 EUROPEAN CONFERENCE ON MOBILE ROBOTS (ECMR), 2019,
  • [4] Continual Vision-based Reinforcement Learning with Group Symmetries
    Liu, Shiqi
    Xu, Mengdi
    Huang, Peide
    Zhang, Xilun
    Liu, Yongkang
    Oguchi, Kentaro
    Zhao, Ding
    [J]. CONFERENCE ON ROBOT LEARNING, VOL 229, 2023, 229
  • [5] Intrinsically Motivated NeuroEvolution for Vision-Based Reinforcement Learning
    Cuccu, Giuseppe
    Luciw, Matthew
    Schmidhuber, Juergen
    Gomez, Faustino
    [J]. 2011 IEEE INTERNATIONAL CONFERENCE ON DEVELOPMENT AND LEARNING (ICDL), 2011,
  • [6] Vision-based Deep Reinforcement Learning to Control a Manipulator
    Kim, Wonchul
    Kim, Taewan
    Lee, Jonggu
    Kim, H. Jin
    [J]. 2017 11TH ASIAN CONTROL CONFERENCE (ASCC), 2017, : 1046 - 1050
  • [7] Review of vision-based reinforcement learning for drone navigation
    Aburaya, Anas
    Selamat, Hazlina
    Muslim, Mohd Taufiq
    [J]. INTERNATIONAL JOURNAL OF INTELLIGENT ROBOTICS AND APPLICATIONS, 2024,
  • [8] Temporal-Spatial Recommendation for Caching at Base Stations via Deep Reinforcement Learning
    Guo, Kaiyang
    Yang, Chenyang
    [J]. IEEE ACCESS, 2019, 7 : 58519 - 58532
  • [9] Approximate Inverse Reinforcement Learning from Vision-based Imitation Learning
    Lee, Keuntaek
    Vlahov, Bogdan
    Gibson, Jason
    Rehg, James M.
    Theodorou, Evangelos A.
    [J]. 2021 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION (ICRA 2021), 2021, : 10793 - 10799
  • [10] Robot Manipulation of Dynamic Object with Vision-based Reinforcement Learning
    Liu, Chenchen
    Zhang, Zhengshen
    Zhou, Lei
    Liu, Zhiyang
    Ang, Marcelo H., Jr.
    Lu, Wenfeng
    Tay, Francis E. H.
    [J]. 2024 9TH INTERNATIONAL CONFERENCE ON CONTROL AND ROBOTICS ENGINEERING, ICCRE 2024, 2024, : 21 - 26