Scholarlink

Unmanned vehicles have a broad application prospect, and path planning is a very important part. Unmanned vehicles need to explore the environment independently to find their destination. The forest land environment is a relatively common field environment. There are many trees in this environment, which requires a higher ability of planning the path of unmanned vehicles. The classic algorithm of Deep Deterministic Policy Gradient (DDPG) has the advantages of greater applicability in continuous control problems in deep reinforcement learning. Aiming at the unmanned vehicles path planning in the forest land environment, this paper established the kinematics model of the vehicle, build the vehicle training environment based on GYM. Based on the depth deterministic strategy gradient (DDPG) algorithm, we designed the continuous state action space, and a new reward function with the help of laser radar information. In order to improve the learning efficiency, the data was normalized, and used the prioritized experience replay. The results show that the algorithm can plan collision-free paths in forest land environment.

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

Unmanned Vehicle Path Planning Based on DDPG Reinforcement Learning Algorithm in Forest Land Environment

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Luhao Li,
  • Yongliang Li,
  • Miaomiao Du,
  • Yongkai Jia,
  • Xuegang Sun,
  • Quanqing Wang

摘要

Unmanned vehicles have a broad application prospect, and path planning is a very important part. Unmanned vehicles need to explore the environment independently to find their destination. The forest land environment is a relatively common field environment. There are many trees in this environment, which requires a higher ability of planning the path of unmanned vehicles. The classic algorithm of Deep Deterministic Policy Gradient (DDPG) has the advantages of greater applicability in continuous control problems in deep reinforcement learning. Aiming at the unmanned vehicles path planning in the forest land environment, this paper established the kinematics model of the vehicle, build the vehicle training environment based on GYM. Based on the depth deterministic strategy gradient (DDPG) algorithm, we designed the continuous state action space, and a new reward function with the help of laser radar information. In order to improve the learning efficiency, the data was normalized, and used the prioritized experience replay. The results show that the algorithm can plan collision-free paths in forest land environment.