H^2MARL: Hierarchical Heterogeneous Multi-agent Reinforcement Learning for Humanoid Robot Control
H^2MARL: Hierarchical Heterogeneous Multi-agent Reinforcement Learning for Humanoid Robot Control
This research paper proposes H^2MARL, a hierarchical heterogeneous multi-agent reinforcement learning framework designed to address the challenges of coordinated whole-body control in humanoid robots. The authors introduce a JDoF-level physical-heterogeneity-aware policy factorization and a closed-loop policy-executor co-optimization mechanism to improve locomotion stability, gait coordination, and motion quality. Experiments conducted in MuJoCo and Isaac Gym demonstrate that H^2MARL outperforms existing homogeneous, coarse-grained, and independently optimized alternatives.
The paper introduces H^2MARL, a hierarchical multi-agent reinforcement learning framework for humanoid robots that addresses physical heterogeneity across joint degrees of freedom and improves locomotion stability through closed-loop co-optimization of high-level planning and low-level execution.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
Source：arXiv Robotics — research abstracts · arxiv.org