基于DDPG+MPC的水稻授粉机器人路径跟踪控制  

Path Tracking Control of Rice Pollination Robot Based on DDPG+MPC

作　　者：温佳 梁喜凤 王永维[2] Wen Jia;Liang Xifeng;Wang Yongwei(College of Mechanical and Electrical Engineering,China Jiliang University,Hangzhou 310018,China;College of Biosystems Engineering and Food Science,Zhejiang University,Hangzhou 310058,China)

机构地区：[1]中国计量大学机电工程学院,杭州310018 [2]浙江大学生物系统工程与食品科学学院,杭州310058

出　　处：《农机化研究》2025年第6期18-25,共8页Journal of Agricultural Mechanization Research

基　　金：国家自然科学基金项目(31971796)。

摘　　要：为提高水稻授粉机器人自主作业路径跟踪的控制精度,提出了一种强化学习DDPG结合模型预测控制MPC的混合控制算法。以DDPG+MPC混合控制框架为基础,通过获取当前时刻状态,以参考路径和当前行驶路径偏差为参照,预测下一时刻机器人状态输出,并利用强化学习DDPG完成对原始MPC控制算法的前轮偏角和加速度补偿,提高路径跟踪的行驶精度,实现授粉机器人按照参考路径高精度自主行驶。仿真验证结果表明:在直线路径下,航向偏差稳定后控制在0.03°以内,改进后DDPG+MPC混合控制算法的横向误差相比于MPC算法降低了0.0014 m,跟踪精度提高了5.7%;在转弯曲线路径下,航向偏差小于0.5°,DDPG+MPC混合控制算法的横向误差相较于MPC算法降低了0.0448 m,跟踪精度提高了151.9%,且在直线进入弯道时调整实时性更快,满足授粉机器人自主作业高精度路径跟踪控制要求。In order to improve the control accuracy of autonomous path tracking of rice pollination robot,this paper proposed a hybrid control algorithm based on reinforcement learning DDPG combined with model predictive control MPC.Based on the DDPG+MPC hybrid control framework,the robot state output at the next moment was predicted by obtaining the current state and referring to the deviation between the reference path and the current driving path.The reinforcement learning DDPG was used to complete the front wheel deflection angle and acceleration compensation of the original MPC control algorithm,so as to improve the driving accuracy of the path tracking and realize the high-precision autonomous driving of the pollination robot according to the reference path.The simulation venfication results show that in the straight path,the heading deviation was controlled within 0.03°after the heading deviation was stable.In terms of average lateral error,the improved DDPG+MPC hybrid algorithm reduced the lateral error by 0.0014 m compared with the MPC algorithm,and the tracking accuracy was improved by 5.7%.Under the turning curve path,the heading deviation was less than 0.5°.The lateral error of the DDPG+MPC hybrid control algorithm was 0.0448 m lower than that of the MPC algorithm,and the tracking accuracy was improved by 151.9%.When the straight line entered the curve,the real-time adjustment was faster,which satisfied the high-precision path tracking control of the autonomous operation of the pollination robot.

关 键 词：杂交水稻 授粉机器人 路径跟踪 模型预测控制 强化学习 

分 类 号：S24[农业科学—农业电气化与自动化]