Simulation of Robot Autonomous Obstacle Avoidance Algorithm Based on Hierarchical Reinforcement Learning
The intelligent robot can perceive the environment information in real-time and control the action traj-ectory by drawing the environment map.In order to improve the path planning ability and reduce the collision proba-bility between the robot and obstacles,this paper proposed an algorithm of autonomous obstacle avoidance based on hierarchical reinforcement learning algorithm.Combined with the moving speed and angular speed,a kinematics model was built,and then local and global coordinate systems were established respectively.Through coordinate transforma-tion,the information of the robot and obstacle was collected.Meanwhile,the hierarchical reinforcement learning archi-tecture was constructed,including three levels:environment information interaction,sub-task selection and root task cooperation.After that,the Q-learning method was used as a reinforcement learning strategy,and the rule of updating the Q-function value was determined.Moreover,environment state information was expressed in the form of Cartesian product.And a reasonable reward function was chosen to improve the learning efficiency.Finally,the optimal action of the robot was controlled by maximizing the Q value.Thus,autonomous obstacle avoidance was achieved.Experimental results show that the proposed method can avoid static and dynamic obstacles accurately,with low computational com-plexity,and it also avoids falling into local optimization.
RobotHierarchical reinforcement learningAutonomous obstacle avoidanceLearning strategiesReward function
NETL
NSTL
万方数据
