基于前视声呐信息的AUV避碰规划研究

在复杂海洋环境中,利用前视声呐获取的障碍物信息指导自治水下机器人（AUV）进行局部避碰。主要采用强化学习的方法对AUV进行控制和决策,综合Q学习算法、BP神经网络和人工势场法对AUV进行避碰规划。强化学习的方法强调AUV在环境的影响中学习,通过环境对不同行为的评价性反馈信号来改变行为选择策略。并且在环境发生变化时,AUV通过学习来实现对新环境的适应,不断改进其自治能力,进而实现在不确定环境下的避障任务。开发了AUV运动规划的虚拟仿真软件系统,仿真实验证明了算法的合理性与可行性。

Research on Obstacle Avoidance Planning for AUV Based on Forward Looking Sonar

The obstacle information obtained with forward sonar is to instruct local avoidance for Autonomous Underwater Vehicles(AUV) in complex sea environment. The reinforcement learning is adopted to control and decision for AUV,and Q-learning,BP neural net,artificial potential is integrated to avoidance planning for AUV. The reinforcement learning method emphasizes that AUV can change behavior select strategy by estimate feedback signal from different environment during AUV learns from environment. When environment changes,AUV can improve autonomous ability by adapting to the new environment and complete obstacle avoidance under uncertain environment. The visual simulation soft system for motion planning and simulation of AUV was developed,and the validity and feasibility were verified through simulation test.