摘要
针对松果采摘机器人移动过程中存在的振动问题,提出一种基于强化学习的指定不敏感(SI)输入整形器控制方法,设计了一种机械手臂击打式的松果采摘系统。首先通过测量运动过程中的振动幅度大小,并给予强化学习代理体不同的奖励函数;接着令代理体寻找最大奖励函数,即达到振动幅度最小的SI输入整形最优参数。然后引入衰减函数使代理体跳出局部最优参数,得到全局最优参数;最后通过仿真实验进一步验证该强化学习的SI输入整形器算法的消振效果。与SI输入整形算法相比,振动幅度降低40%,时滞时间缩短46.3 ms。结果表明,所提出的应用强化学习的SI输入整形器具有较好的振动抑制效果和鲁棒性。
According to the vibration problem in the moving of pine nut picking robot,by SI input shaping controlling method on account of reinforcement learning,we designed a pine nut picking robot working by striking type robotic arm.At first,by measuring the vibration amplitude in motion,give the reinforcement learning agent different reward function.Next,let the agent find the Maximum reward function which is the SI input shaping optimal parameter that appears when vibration amplitude to minimum.Then,leading in attenuation function,making the agent jump out of local optimal parameter,and getting the global optimal parameter.Finally,by simulation experiment,further test and verity the reinforcement learning SI input shaping’s vibration elimination effect.Compared with SI input shaping algorithm,the vibration amplitude lowered by 40%,and the skew-time shortened by 46.3 ms.The SI input shaping with the reinforcement learning has a better vibration suppression effect and robustness.
作者
徐健
曹军
张怡卓
Xu Jian;Cao Jun;Zhang Yizhuo(Northeast Forestry University,Harbin 150040,P.R.China)
出处
《东北林业大学学报》
CAS
CSCD
北大核心
2020年第12期79-84,共6页
Journal of Northeast Forestry University
基金
林业公益性行业科研专项(201504307)。
关键词
输入整形
强化学习
参数优化
机械臂
Input shaping
Reinforcement learning
Parameter optimization
Robotic arms
