Scholarlink

With the rise of the low-altitude economy, multi-Unmanned Aerial Vehicle (UAV) coordination for tasks like path tracking and collision avoidance has gained prominence. Existing discrete diffusion Soft Actor-Critic (DSAC) methods do not meet the requirements of continuous action spaces. Therefore, we propose a continuous DSAC approach with three variance control strategies: diffusion-generated variance, self-learned parameters, and fixed-variance to address this limitation. We evaluated them in a 3D spiral trajectory tracking task with artificial potential field dynamics. Experiments demonstrate that self-learned parameters outperform alternatives and achieve higher rewards and better coordination. This work provides a promising approach to adaptive UAV control in low-altitude applications.

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

Variance-Controlled Diffusion SAC for Multi-UAV Coordination in Low-Altitude Environments

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Fuzheng Guo,
  • Qianjin Li

摘要

With the rise of the low-altitude economy, multi-Unmanned Aerial Vehicle (UAV) coordination for tasks like path tracking and collision avoidance has gained prominence. Existing discrete diffusion Soft Actor-Critic (DSAC) methods do not meet the requirements of continuous action spaces. Therefore, we propose a continuous DSAC approach with three variance control strategies: diffusion-generated variance, self-learned parameters, and fixed-variance to address this limitation. We evaluated them in a 3D spiral trajectory tracking task with artificial potential field dynamics. Experiments demonstrate that self-learned parameters outperform alternatives and achieve higher rewards and better coordination. This work provides a promising approach to adaptive UAV control in low-altitude applications.