English

Bias-reduced Multi-step Hindsight Experience Replay for Efficient Multi-goal Reinforcement Learning

Machine Learning 2022-09-27 v3 Artificial Intelligence Robotics

Abstract

Multi-goal reinforcement learning is widely applied in planning and robot manipulation. Two main challenges in multi-goal reinforcement learning are sparse rewards and sample inefficiency. Hindsight Experience Replay (HER) aims to tackle the two challenges via goal relabeling. However, HER-related works still need millions of samples and a huge computation. In this paper, we propose Multi-step Hindsight Experience Replay (MHER), incorporating multi-step relabeled returns based on nn-step relabeling to improve sample efficiency. Despite the advantages of nn-step relabeling, we theoretically and experimentally prove the off-policy nn-step bias introduced by nn-step relabeling may lead to poor performance in many environments. To address the above issue, two bias-reduced MHER algorithms, MHER(λ\lambda) and Model-based MHER (MMHER) are presented. MHER(λ\lambda) exploits the λ\lambda return while MMHER benefits from model-based value expansions. Experimental results on numerous multi-goal robotic tasks show that our solutions can successfully alleviate off-policy nn-step bias and achieve significantly higher sample efficiency than HER and Curriculum-guided HER with little additional computation beyond HER.

Keywords

Cite

@article{arxiv.2102.12962,
  title  = {Bias-reduced Multi-step Hindsight Experience Replay for Efficient Multi-goal Reinforcement Learning},
  author = {Rui Yang and Jiafei Lyu and Yu Yang and Jiangpeng Yan and Feng Luo and Dijun Luo and Lanqing Li and Xiu Li},
  journal= {arXiv preprint arXiv:2102.12962},
  year   = {2022}
}

Comments

20pages, 8 figures

R2 v1 2026-06-23T23:30:49.907Z