Interpretable Reward Redistribution in Reinforcement Learning: A Causal Approach

被引:0
|
作者
Zhang, Yudi [1 ]
Du, Yali [2 ]
Huang, Biwei [3 ]
Wang, Ziyan [2 ]
Wang, Jun [4 ]
Fang, Meng [1 ,5 ]
Pechenizkiy, Mykola [1 ]
机构
[1] Eindhoven Univ Technol, Eindhoven, Netherlands
[2] Kings Coll London, London, England
[3] Univ Calif San Diego, La Jolla, CA 92093 USA
[4] UCL, London, England
[5] Univ Liverpool, Liverpool, Merseyside, England
基金
英国工程与自然科学研究理事会;
关键词
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
A major challenge in reinforcement learning is to determine which state-action pairs are responsible for future rewards that are delayed. Reward redistribution serves as a solution to re-assign credits for each time step from observed sequences. While the majority of current approaches construct the reward redistribution in an uninterpretable manner, we propose to explicitly model the contributions of state and action from a causal perspective, resulting in an interpretable reward redistribution and preserving policy invariance. In this paper, we start by studying the role of causal generative models in reward redistribution by characterizing the generation of Markovian rewards and trajectory-wise long-term return and further propose a framework, called Generative Return Decomposition (GRD), for policy optimization in delayed reward scenarios. Specifically, GRD first identifies the unobservable Markovian rewards and causal relations in the generative process. Then, GRD makes use of the identified causal generative model to form a compact representation to train policy over the most favorable subspace of the state space of the agent. Theoretically, we show that the unobservable Markovian reward function is identifiable, as well as the underlying causal structure and causal models. Experimental results show that our method outperforms state-of-the-art methods and the provided visualization further demonstrates the interpretability of our method. The project page is located at https://reedzyd.github.io/GenerativeReturnDecomposition/.
引用
收藏
页数:22
相关论文
共 50 条
  • [1] Reward Redistribution for Reinforcement Learning of Dynamic Nonprehensile Manipulation
    Sejnova, Gabriela
    Mejdrechova, Megi
    Otahal, Marek
    Sokovnin, Nikita
    Farkas, Igor
    Vavrecka, Michal
    [J]. 2021 7TH INTERNATIONAL CONFERENCE ON CONTROL, AUTOMATION AND ROBOTICS (ICCAR), 2021, : 326 - 331
  • [2] Leveraging Reward Consistency for Interpretable Feature Discovery in Reinforcement Learning
    Yang, Qisen
    Wang, Huanqian
    Tong, Mukun
    Shi, Wenjie
    Huang, Gao
    Song, Shiji
    [J]. IEEE TRANSACTIONS ON SYSTEMS MAN CYBERNETICS-SYSTEMS, 2024, 54 (02): : 1014 - 1025
  • [3] Generation of Roles in Reinforcement Learning Considering Redistribution of Reward between Agents
    Nakahara, Masayuki
    Osana, Yuko
    [J]. 2009 IEEE INTERNATIONAL CONFERENCE ON SYSTEMS, MAN AND CYBERNETICS (SMC 2009), VOLS 1-9, 2009, : 2259 - +
  • [4] Exploiting Transformer in Sparse Reward Reinforcement Learning for Interpretable Temporal Logic Motion Planning
    Zhang, Hao
    Wang, Hao
    Kan, Zhen
    [J]. IEEE ROBOTICS AND AUTOMATION LETTERS, 2023, 8 (08) : 4831 - 4838
  • [5] A formal methods approach to interpretable reinforcement learning for robotic planning
    Li, Xiao
    Serlin, Zachary
    Yang, Guang
    Belta, Calin
    [J]. SCIENCE ROBOTICS, 2019, 4 (37)
  • [6] Reward tampering problems and solutions in reinforcement learning: a causal influence diagram perspective
    Everitt, Tom
    Hutter, Marcus
    Kumar, Ramana
    Krakovna, Victoria
    [J]. SYNTHESE, 2021, 198 (SUPPL 27) : 6435 - 6467
  • [7] Reward tampering problems and solutions in reinforcement learning: a causal influence diagram perspective
    Tom Everitt
    Marcus Hutter
    Ramana Kumar
    Victoria Krakovna
    [J]. Synthese, 2021, 198 : 6435 - 6467
  • [8] A survey on interpretable reinforcement learning
    Glanois, Claire
    Weng, Paul
    Zimmer, Matthieu
    Li, Dong
    Yang, Tianpei
    Hao, Jianye
    Liu, Wulong
    [J]. MACHINE LEARNING, 2024, 113 (08) : 5847 - 5890
  • [9] Interpretable Control by Reinforcement Learning
    Hein, Daniel
    Limmer, Steffen
    Runkler, Thomas A.
    [J]. IFAC PAPERSONLINE, 2020, 53 (02): : 8082 - 8089
  • [10] Programmatically Interpretable Reinforcement Learning
    Verma, Abhinav
    Murali, Vijayaraghavan
    Singh, Rishabh
    Kohli, Pushmeet
    Chaudhuri, Swarat
    [J]. INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 80, 2018, 80