A deep reinforcement learning path planning method based on local environment driving

Abstract

The application discloses a kind of local environment driving-based deep reinforcement learning path planning method, belong to mobile robot technical field, for the autonomous navigation and obstacle avoidance of mobile robot in complex environment.The method first constructs obstacle profile model according to the obstacle point information in local environment;Then local environment graph structure between obstacle and robot is constructed using undirected graph, spatial features and deep feature information in local environment are extracted using graph attention network;Then the optimal local target point is obtained using local target driving mechanism;Finally, the optimal path planning strategy is trained by fusing deep reinforcement learning.The application guides local target point and enhances the key features in local environment by graph attention network reinforcement learning, improves the local environment perception ability of robot, enhances the path planning ability of robot in complex environment, can significantly shorten the average path length while improving the navigation success rate, enhances trajectory smoothness.

Images