Scholarlink

To address the challenges of safe obstacle avoidance and stable formation for multiple robots in complex dynamic environments, this paper proposes a hierarchical reinforcement learning scheme based on TD3(Twin Delayed Deep Deterministic Policy Gradient)( \(HRL-TD3\) ). Given the limited environmental adaptability of traditional methods, the scheme constructs a perception-decision loop through a multi-head neural network: the upper layer parses lidar information to generate strategies, while the lower layer deploys scenario-optimized TD3 sub-models. Experimental results show that this scheme outperforms the virtual spring method(VSM) and the improved artificial potential field method(APF) in terms of efficiency, arrival time, and scalability. The collision probability of robot formations is below 11%, which can meet the needs of large-scale systems.

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

Research on Obstacle Avoidance Control Method for Unmanned Vehicle Formation Based on Hierarchical Reinforcement Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Hang Zhang,
  • Jie Huang,
  • Nan Wang,
  • Huajie Hong,
  • Lei Ding

摘要

To address the challenges of safe obstacle avoidance and stable formation for multiple robots in complex dynamic environments, this paper proposes a hierarchical reinforcement learning scheme based on TD3(Twin Delayed Deep Deterministic Policy Gradient)( \(HRL-TD3\) ). Given the limited environmental adaptability of traditional methods, the scheme constructs a perception-decision loop through a multi-head neural network: the upper layer parses lidar information to generate strategies, while the lower layer deploys scenario-optimized TD3 sub-models. Experimental results show that this scheme outperforms the virtual spring method(VSM) and the improved artificial potential field method(APF) in terms of efficiency, arrival time, and scalability. The collision probability of robot formations is below 11%, which can meet the needs of large-scale systems.