Inspection robot navigation method based on deep reinforcement learning

The invention discloses an inspection robot navigation method based on deep reinforcement learning, and relates to the technical field of security inspection intelligence. According to the invention, the robot inspection environment is sensed only by using specially processed laser radar data and odometer data, compared with depth camera and radar fusion data, the hardware calculation cost is reduced, the difference with the real environment is also reduced, and a navigation decision model is easier to deploy and higher in applicability; the invention provides an improved Actor and Critic network, an LSTM layer is added in the Actor and Critic network, the relationship among action information of recent several time steps is extracted, and the final execution action is optimized, so that the moving action of the inspection robot is smoother, and a smoother navigation path is generated. 本发明公开了一种基于深度强化学习的巡检机器人导航方法，涉及安防巡检智能化技术领域。本发明仅利用经过特殊处理的激光雷达数据和里程计数据对机器人巡检环境进行感知，与基于深度相机与雷达融合数据相比，减少硬件计算成本的同时也减少了与现实环境的差距，使导航决策