计算机科学
运动规划
水下
路径(计算)
卷积神经网络
过程(计算)
人工智能
避障
障碍物
计算机视觉
实时计算
人工神经网络
机器视觉
机器人
移动机器人
海洋学
政治学
法学
程序设计语言
地质学
操作系统
作者
Yibo Wang,Xiaofeng Meng,Yubo Weng
摘要
Autonomous Underwater Vehicle (AUV) is an important tool for intelligent ocean applications, which can be applied to detect underwater environment and search target. Path planning is the key technology to realize AUV intelligence, which has important research significance. Aiming at the dynamic path planning problem, the Regional Ocean Modeling System (ROMS) was first applied to the AUV three-dimensional (3D) dynamic path planning, and a Gate Recurrent Unit Proximal Policy Optimization with Local Vision (GPPO-LV) model based on local vision is proposed. The 3D ocean current environment is constructed based on the ROMS simulation data. The local vision matrix is constructed based on underwater images, and the features of local vision are extracted using convolutional neural network. Furthermore, Gate Recurrent Unit (GRU) network is used to mine the hidden information between observation states. Finally, the Actor network for strategy output and the Critical network for action value evaluation are constructed, and strategies are optimized in the process of interaction with the environment. The experiment shows that under various unknown environments, AUV can carry out real-time path planning under real ocean current data, and has good obstacle avoidance ability.
科研通智能强力驱动
Strongly Powered by AbleSci AI