Scholarlink

To achieve sustained autonomous dynamic soaring, trajectory planning is indispensable. Common approaches require accurate knowledge of the environment for planning. This work aims to develop a trajectory planner using reinforcement learning to circumvent the necessity of having perfect knowledge of the environment. To reduce the dimension and the computational complexity of the problem, we propose an efficient trajectory parameterization. Although the planner is trained only for a specific wind speed, it enables steady dynamic soaring across various wind conditions. Eventually, the developed planner is benchmarked against closed-loop parameter optimization and open-loop trajectory optimization approaches.

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

Reinforcement Learning Based Trajectory Planning for Autonomous Dynamic Soaring

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Alexander Zwenig,
  • Samuel Layer-Reiss,
  • Florian Holzapfel,
  • Haichao Hong

摘要

To achieve sustained autonomous dynamic soaring, trajectory planning is indispensable. Common approaches require accurate knowledge of the environment for planning. This work aims to develop a trajectory planner using reinforcement learning to circumvent the necessity of having perfect knowledge of the environment. To reduce the dimension and the computational complexity of the problem, we propose an efficient trajectory parameterization. Although the planner is trained only for a specific wind speed, it enables steady dynamic soaring across various wind conditions. Eventually, the developed planner is benchmarked against closed-loop parameter optimization and open-loop trajectory optimization approaches.