Scholarlink

Hexapod robots offer excellent flexibility, stability, and environmental adaptability. However, their complex structures and operating conditions pose challenges for gait control. To improve its low robustness and poor generalization in control, we propose a deep reinforcement learning approach with domain randomization (DR-DRL) for locomotion control. By systematically randomizing domain parameters, we construct dynamic training environments and train controllers using a distributed LSTM-PPO algorithm. Sim-to-Sim evaluations across diverse platforms show that our method significantly outperforms standard baselines in perturbation resistance, demonstrating improved robustness and transferability of the learned policies.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Reinforcement Learning for Hexapod Robots Gait Learning and Control with Domain Randomization

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Dengfeng Hong,
  • Guofei Xiang,
  • Lun Li,
  • Xingxing You,
  • Songyi Dian

摘要

Hexapod robots offer excellent flexibility, stability, and environmental adaptability. However, their complex structures and operating conditions pose challenges for gait control. To improve its low robustness and poor generalization in control, we propose a deep reinforcement learning approach with domain randomization (DR-DRL) for locomotion control. By systematically randomizing domain parameters, we construct dynamic training environments and train controllers using a distributed LSTM-PPO algorithm. Sim-to-Sim evaluations across diverse platforms show that our method significantly outperforms standard baselines in perturbation resistance, demonstrating improved robustness and transferability of the learned policies.