基于自适应神经模糊网络的果蔬抓取力控制

运用自适应神经模糊推理系统设计了农业机器人果蔬抓取力智能控制器。以当前抓取力和滑觉传感器信号的小波变换细节系数作为控制器的输入,末端执行器两指闭合距离作为控制器的输出。基于减法聚类建立模糊推理模型,通过调整聚类半径来优选模糊规则数。给出了训练样本数据集采集方法,并应用梯度下降与最小二乘混合训练算法辨识了控制器的前件参数和结论参数。对所设计的控制器进行了实验验证,结果表明该控制器能够适应果蔬质量、表面摩擦特性等方面的差异。抓取力超调量得到了限制,最大值小于0.8 N,可以避免给抓取对象造成机械损伤。

Griping Force Control Using Adaptive Neuro-fuzzy Inference Systems

An intelligent controller using adaptive neuro-fuzzy inference system was developed to control the force of gripping fruits and vegetables of an agricultural robot. The inputs of the controller are the current griping force and the detail coefficients of discrete wavelet transform of the signal from slipping sensor fixed on the robotic end effector. The output of the controller is the displacement of fingers of the end effector. Firstly, a subtractive clustering was applied to generate a fuzzy model, and the radius of the clustering was adjusted to optimize the fuzzy rules. Then methods of sampling training data were introduced, and a hybrid training algorithm consisting of the gradient descent and least square algorithms was implemented to tune antecedent parameters and consequent part of the model. Finally, the experiments of controlling the griping force were carried out. It shows that the controller is able to adapt itself to differences of the fruits and vegetables in mass and surface friction characteristics. Moreover the controlling overshoot of griping force is restrained successfully and less than 0.8N, which prevented the grasping of fruits and vegetables from mechanical destruction.