Scholarlink

In the field of indoor rescue, the indoor guidance technology of UAV has been applied more and more. In large buildings, the indoor anti-collision problem of drones is particularly critical. There are few obstacle classification studies on UAV flight collision avoidance strategies. Accurate Classification DQN (ACDQN) policy framework is proposed. Based on the types of obstacles, the classification is carried out, and the optimal strategy is obtained based on the surrounding environment information to improve the flight safety performance. After the classification of obstacles, the corresponding evaluation function is made according to the actual level, and the safety evaluation is carried out according to the randomness of the obstacles and the distance between them and the UAV. Finally, the collision prediction module is integrated to improve the flight stability performance. The final model is obtained by using the parameters analyzed by ACDQN framework. The results show that ACDQN framework has higher security and stability.

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

Contribution Research on Collision Avoidance Method of Indoor UAV Rescue Based on Reinforcement Learning

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Chengcheng Zhang,
  • Xin Wang,
  • Zhijun Gao,
  • Zhendong Gao,
  • Fei Xiao,
  • Wenping Jiang

摘要

In the field of indoor rescue, the indoor guidance technology of UAV has been applied more and more. In large buildings, the indoor anti-collision problem of drones is particularly critical. There are few obstacle classification studies on UAV flight collision avoidance strategies. Accurate Classification DQN (ACDQN) policy framework is proposed. Based on the types of obstacles, the classification is carried out, and the optimal strategy is obtained based on the surrounding environment information to improve the flight safety performance. After the classification of obstacles, the corresponding evaluation function is made according to the actual level, and the safety evaluation is carried out according to the randomness of the obstacles and the distance between them and the UAV. Finally, the collision prediction module is integrated to improve the flight stability performance. The final model is obtained by using the parameters analyzed by ACDQN framework. The results show that ACDQN framework has higher security and stability.