【作者】 周治国； 何栩； 徐力生； 屈崇；

【Author】 Zhiguo ZHOU;Xu HE;Lisheng XU;Chong QU;School of Information and Electronics,Beijing Institute of Technology;Shanghai Marine Diesel Engine;

【机构】 北京理工大学信息与电子学院； 中国船舶集团有限公司第七一一研究所；

【摘要】 安全无碰撞的自主导航是无人艇运行的基础。大多无人艇路径规划仿真与实船实验差距大,路径规划方法对新环境的适应性不足。针对这些问题,使用Unity3D搭建高保真度的无人艇路径规划仿真平台,使用虚拟激光雷达获取局部环境信息,通过船舶运动数学模型仿真无人艇运动;实现适用于无人艇的LSTM-PPO深度强化学习路径规划算法,设计适合无人艇的奖励函数,使得无人艇能够规避静态与动态障碍物并抵达目的地。在多个复杂地图中进行对比实验,实验结果表明,仿真具有较好的真实度和视觉效果,算法能够实现无人艇安全行驶,且避障能力和面对新环境的适应性较强。更多还原

【Abstract】 Safe and collision-free navigation is the basis of Unmanned Surface Vehicle(USV) functioning.There exists a large gap between most USV path planning simulations and real ship experiments,and most path planning methods are not adaptable to new environments.To solve these problems,a high-fidelity USV path planning simulation platform was built using Unity3D.Virtual lidar was used to obtain local environmental information,and USV motion can be simulated through mathematical model of ship motion.A USV-oriented LSTM-PPO deep reinforcement learning path planning algorithm was realized,and the reward function specialized for USV was designed,so that USV can avoid static and moving obstacles and reach the destination.Comparative experiments were carried out on multiple complex maps.The experimental results show that the simulation is of authenticity and has good visual effects.The algorithm can achieve safe navigation of USV,and shows excellent ability to avoid obstacles and adapt to new environments.更多还原