双指型农业机器人抓取球形果蔬的控制器设计

Controller design for realizing double-finger agricultural robot to grasp spherical fruits and vegetables

现阶段针对苹果、番茄这类球形果蔬设计的采摘机器人末端执行器大多采用相似的双指结构,但是这类采摘机器人的抓取控制方法均是一对一的研究,研发成本偏高且产品不具有通用性。阻抗控制是常用的柔顺控制方法,能够将果蔬的受力-形变等效为环境导纳模型,不同的果蔬对应不同的导纳模型,但是基本的阻抗控制框架是一致的。因此,本文分别选取偏软和偏硬的两种球形果实,通过压缩试验模拟抓取过程,得到果实受力及形变数据,并以此计算得到果实等效刚度。接着,使用BP神经网络分别对原始数据进行训练及验证,得到各层权值。然后在Simulink中建立环境导纳模型并根据不同的果实选择不同的权值,并完成阻抗控制系统的搭建。同时,根据环境导纳模型中的等效刚度估计值实现在线调整阻抗控制器刚度参数。最后通过Matlab进行仿真验证,结果表明改进后方法对于抓取苹果和番茄两种果实,期望力超调分别为2.2%和1.5%,位置控制器输出最终分别稳定在0.55 mm和4.02 mm范围内,可验证所提方法在理论上能够实现农业机器人的柔性抓取。

Most of the picking robot end effectors designed for spherical fruits and vegetables such as apples and tomatoes adopt similar double-finger structures currently.However,the grasping control methods of such picking robots are one-to-one research,the development costs are high and the products have no generality.Impedance control as a commonly used compliant control method can equalize the force-deformation of fruits and vegetables to the environmental admittance model.Different fruits and vegetables correspond to different admittance models,but the basic impedance control framework is consistent.Therefore,two different kinds of spherical fruits,which are soft and hard,were selected in this paper.The compression process was used to simulate the grasping process,the fruit force-deformation data were obtained,and the equivalent stiffness of the fruit was calculated.Then,the BP neural network is used to train and verify the original data to obtain the weights of each layer.Establish the environmental admittance model in Simulink and choose different weight values according to different fruits,and the construction of the impedance control system is completed.At the same time,the stiffness parameter of the impedance controller is adjusted online according to the equivalent stiffness estimation value in the environmental admittance model.Finally,by the Matlab simulation experiments,the results showed that for the two fruits of apple and tomato,the expected overshoots of the improved method are 2.2%and 1.5%,respectively.The position controller outputs are stable in the range of 0.55 mm and 4.02 mm respectively.It can be verified that the proposed method can theoretically realize the flexible grasping of agricultural robots.