AUV Path Planning and Object Tracking Algorithm Based on Reinforcement Learning Under Point Cloud Observation

Meng Xi, Linfei Cao, Jiabao Wen, Zhijing Wang, Qi Wu, Jia Wang

Abstract: Autonomous underwater vehicles (AUV) play an important role in the process of human exploration of the ocean. However, the existing AUV control methods are faced with the problem of insufficient intelligence and can not adapt to the complex marine environment. Deep reinforcement learning approximates nonlinear functions with deep neural networks. Applying deep reinforcement learning to the path planning of AUV can solve the complex situation that traditional control methods cannot handle. We study the path planning and target tracking of AUV. Two AUVs were trained to complete the piloting and following tasks respectively. The piloting AUV could detect the obstacle information in the surrounding environment and reach the target point while completing the obstacle avoidance task. The mission of the following AUV is to stay at a distance from the leader to form an AUV swarm. We built a marine environment in the gazebo, which is a relatively realistic physics engine,and added sensors on the AUVs to obtain the information of the surrounding environment. By using soft actor critic algorithm, we realized the AUV cluster control, which can simultaneously complete the tasks of path planning, obstacle avoidance and target tracking.

External IDs:doi:10.1109/tce.2025.3614074