arXiv Robotics — research abstracts· Yonghoon Dong, Kyungmin Lee, Changyeon Kim, Jaehyuk Kim, Jinwoo Shin·· 1 days agoEditorial score42
Trust Region Q Adjoint Matching for Stable Off-Policy RL
Trust Region Q Adjoint Matching
Summary
TRQAM improves off-policy reinforcement learning stability by controlling path-space KL divergence via a trust-region parameter. It outperforms prior methods with 68% success in offline RL.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
Source：arXiv Robotics — research abstracts · arxiv.org