基于Q-学习的果园机器人避障算法研究

Research on Obstacle Avoidance Algorithms for Robot Based on Q-Learning

针对果园环境复杂、自动化程度低的问题,提出了基于Q-学习的果园机器人避障算法。该果园机器人搭载了激光雷达传感器。为了降低激光雷达数据的冗余度,首先对激光雷达数据进行截取,保留了机器人前进方向上一定角度内的数据,再将激光雷达的数据栅格化,采用数据差的方式标记出障碍物,将所有障碍物补齐为规定的正方形;然后,建立果园机器人的差速模型,对Q-学习的R值表、动作空间和动作选择策略进行建模,使得果园机器人能在果园路径规划过程中所获奖赏最大。MatLab仿真结果表明:该方法能够规划出一条较优路径。室内试验表明:机器人能平稳地避开所设障碍物,到达指定终点。

To solve the problem of complex orchard environment and low degree of mechanization,this paper proposes an obstacle avoidance algorithm for robot based on Q-learning.The orchard robot is equipped with a lidar sensor.In order to reduce the redundancy of lidar data,firstly,the data of lidar is filtered,and the data in a certain angle of the forward direction is retained.After rasterizing the data,the obstacles are marked by the difference of data,and all obstacles are marked square,to build the differential model of the robot.Then the R-value table,action space and moving selection strategy of Q-learning are modeled so that the robot can have the greatest feedback in the route planning.The simulation results by MATLAB show that the method can plan a better route.Finally,the indoor experiments show that the robot can successfully avoid the obstacles and reach the designated.This method provides a basis for route planning of orchard robot.