摘要
针对智能除草机器人领域缺少主动避苗路径规划问题,提出了一种基于改进的深度确定性策略梯度除草机器人机械臂路径规划算法。通过引入奖励等势面的概念改进DDPG算法,并利用CoppeliaSim软件搭建仿真训练环境,对提出的算法进行训练与验证。研究表明:改进的DDPG算法在仿真环境下除草成功率为93.36%,伤苗率为2.79%。同时,通过搭建测试平台,在实际温室环境下进行了田间除草试验,试验结果与仿真结果一致,实际除草成功率为91.50%、伤苗率为2.82%。研究结果表明:所提出的算法能够在实际环境中有效减少除草机器人除草作业时对作物幼苗的损伤。
The current state of research on active seedling avoidance path planning in the field of intelligent weeding robots is inadequate.To address this issue,a new path planning algorithm was developed for gradient weeding robot manipulators,utilizing an improved depth deterministic strategy.The algorithm was enhanced through the incorporation of reward equipotential surfaces,which improved the performance of the DDPG algorithm.To validate the algorithm,a simulation training environment was constructed using CoppeliaSim software,where the algorithm was trained and verified.The results showed 93.36%success rate for weeding and 2.79%rate of seed injury.A test platform was built to conduct weeding tests in an actual environment,where the algorithm achieved 91.50%success rate for weeding and 2.82%rate of seed injury.These experimental findings demonstrated the efficacy of the proposed algorithm in reducing the damage to crop.
作者
杨卜
邬鑫
张梦磊
冯松科
Yang Bu;Wu Xin;Zhang Menglei;Feng Songke(Collage of Mechanical and Electronic Engineering,Northwest A&F University,Yangling 712100,China)
出处
《农机化研究》
北大核心
2025年第5期15-21,共7页
Journal of Agricultural Mechanization Research
基金
国家青年自然科学基金项目(51804260)。
关键词
除草机器人
深度强化学习
路径规划
人工势场法
weeding robot
deep reinforcement learning
path planning
artificial potential field method
