Stein Variational Goal Generation for adaptive Exploration in Multi-Goal Reinforcement Learning

被引:0
|
作者
Castanet, Nicolas [1 ]
Sigaud, Olivier [1 ]
Lamprier, Sylvain [2 ]
机构
[1] Sorbonne Univ, ISIR, Paris, France
[2] Univ Angers, LERIA, SFR MATHSTIC, F-49000 Angers, France
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
In multi-goal Reinforcement Learning, an agent can share experience between related training tasks, resulting in better generalization for new tasks at test time. However, when the goal space has discontinuities and the reward is sparse, a majority of goals are difficult to reach. In this context, a curriculum over goals helps agents learn by adapting training tasks to their current capabilities. In this work we propose Stein Variational Goal Generation (SVGG), which samples goals of intermediate difficulty for the agent, by leveraging a learned predictive model of its goal reaching capabilities. The distribution of goals is modeled with particles that are attracted in areas of appropriate difficulty using Stein Variational Gradient Descent. We show that SVGG outperforms state-of-the-art multi-goal Reinforcement Learning methods in terms of success coverage in hard exploration problems, and demonstrate that it is endowed with a useful recovery property when the environment changes.
引用
收藏
页数:18
相关论文
共 50 条
  • [1] Guided goal generation for hindsight multi-goal reinforcement learning
    Bai, Chenjia
    Liu, Peng
    Zhao, Wei
    Tang, Xianglong
    NEUROCOMPUTING, 2019, 359 : 353 - 367
  • [2] Adaptive Multi-Goal Exploration
    Tarbouriech, Jean
    Domingues, Omar Darwiche
    Menard, Pierre
    Pirotta, Matteo
    Valko, Michal
    Lazaric, Alessandro
    INTERNATIONAL CONFERENCE ON ARTIFICIAL INTELLIGENCE AND STATISTICS, VOL 151, 2022, 151
  • [3] Overfitting-avoiding goal-guided exploration for hard-exploration multi-goal reinforcement learning
    Han, Changlin
    Peng, Zhiyong
    Liu, Yadong
    Tang, Jingsheng
    Yu, Yang
    Zhou, Zongtan
    NEUROCOMPUTING, 2023, 525 : 76 - 87
  • [4] Combining Hindsight with Goal-enhanced Prediction for Multi-goal Reinforcement Learning
    Yang, Rui
    Luo, Feng
    Li, Xiu
    2021 IEEE 33RD INTERNATIONAL CONFERENCE ON TOOLS WITH ARTIFICIAL INTELLIGENCE (ICTAI 2021), 2021, : 314 - 321
  • [5] Maximum Entropy-Regularized Multi-Goal Reinforcement Learning
    Zhao, Rui
    Sun, Xudong
    Tresp, Volker
    INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 97, 2019, 97
  • [6] Multi-goal Reinforcement Learning via Exploring Successor Matching
    Feng, Xiaoyun
    2022 IEEE CONFERENCE ON GAMES, COG, 2022, : 401 - 408
  • [7] CURIOUS: Intrinsically Motivated Modular Multi-Goal Reinforcement Learning
    Colas, Cedric
    Fournier, Pierre
    Sigaud, Olivier
    Chetouani, Mohamed
    Oudeyer, Pierre-Yves
    INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 97, 2019, 97
  • [8] Hierarchical reinforcement learning for handling sparse rewards in multi-goal navigation
    Yan, Jiangyue
    Luo, Biao
    Xu, Xiaodong
    ARTIFICIAL INTELLIGENCE REVIEW, 2024, 57 (06)
  • [9] Efficient Multi-Goal Reinforcement Learning via Value Consistency Prioritization
    Xu, Jiawei
    Li, Shuxing
    Yang, Rui
    Yuan, Chun
    Han, Lei
    JOURNAL OF ARTIFICIAL INTELLIGENCE RESEARCH, 2023, 77 : 355 - 376
  • [10] Efficient Multi-Goal Reinforcement Learning via Value Consistency Prioritization
    Xu J.
    Li S.
    Yang R.
    Yuan C.
    Han L.
    Journal of Artificial Intelligence Research, 2023, 77 : 355 - 376