基于卷积神经网络的机械臂抓取控制系统设计

Design of Manipulator Grasping Control System Based on Convolutional Neural Network

摘要:

为保证机械臂的抓取精度，保证物体抓取的稳定性，本文设计基于卷积神经网络的机械臂抓取控制系统。在系统硬件部分，加设图像、位置和压力传感器，改装机械臂抓取控制器和运动驱动器，利用图像传感器设备，获取满足质量要求的机械臂抓取目标图像，为机械臂抓取控制功能提供硬件支持。软件部分利用卷积神经网络算法提取图像特征，确定机械臂抓取目标位置。结合机械臂当前位置的检测结果，规划机械臂抓取路线，预估机械臂抓取角度与抓取力。最终通过机械臂抓取参数控制量的计算，在控制器的支持下实现系统的机械臂抓取控制功能。实验结果表明，所设计系统应用下位置控制误差和速度控制误差的平均值分别为0.192m和0.138m/s，同时物体抓取掉落概率明显降低。

Abstract:

In order to ensure the grasping accuracy of the manipulator and the stability of object grasping, this paper designs a manipulator grasping control system based on convolutional neural network. In the hardware part of the system, image, position and pressure sensors are added, the manipulator grasping controller and motion driver are modified, and the image sensor equipment is used to obtain the manipulator grasping target image that meets the quality requirements, providing hardware support for the manipulator grasping control function. In the software part, the convolution neural network algorithm is used to extract the image features and determine the position of the target grasped by the manipulator. Based on the detection results of the current position of the manipulator, the grasping route of the manipulator is planned, and the grasping angle and grasping force of the manipulator are estimated. Finally, through the calculation of the manipulator grasping parameter control quantity, the manipulator grasping control function of the system is realized under the support of the controller. The experimental results show that the average values of position control error and speed control error are 0.192 m and 0.138 m/s respectively, and the probability of object grasping and falling is significantly reduced.