基于视觉测量与神经网络的工业机器人位姿补偿

为提高六轴工业机器人的绝对定位精度，本文提出了一种利用视觉测量数据通过ELM(Extreme Learning Machine)神经网络实现机器人位姿补偿的新方法．利用固定在机器人末端的手眼相机获取机器人的末端位姿，并借助ELM实现机器人末端执行器从目标位姿到预测指令位姿之间映射，用修正转角代替原转角使机器人末端执行器运行至修正位姿，实现补偿．特别的是，对于使用的ELM，以网络预测均方误差为指标定量选取了网络的最佳参数．相比之前的方法，本文提出的算法具有能够同时高精度补偿姿态角及位置误差的显著优点．为验证该位姿误差补偿方法的有效性，本文进行了实验验证．结果表明，相比较于未补偿前的机器人末端位姿误差，经该方法补偿后的位姿误差被稳定控制在较低水平，平均位置误差降低89.1 %；平均姿态角误差降低96.8 %．除此以外，位置误差与姿态角误差的标准差也分别降低了85.66 %和93.24 %．

Industrial Robot Pose Compensation Based on Vision-Based Measurement and Neural Network

In order to improve the absolute positioning accuracy of the six-axis industrial robot, a new robot compensation method using vision-based measurement and ELM(Extreme Learning Machine) neural network is proposed. The hand-eye camera fixed at the end of the robot is used to obtain the pose of the robot. The target pose is mapped to the predicted command pose with the help of ELM, and the corrected joint angles are used instead of the original joint angles to command the robot end effector to the corrected pose to achieve compensation. In particular, for the ELM used, the optimal parameters of the network are quantitatively selected with the mean square error of network prediction as the index. Compared with previous methods, the proposed method is more advanced as it can compensate not only robot end positions but also the orientation angles. Experiments are carried out in order to verify the effectiveness of the proposed method. The results show that, compared with the error of the robot end pose before compensation, the error after compensation is stably controlled at a lower level, with the average position error reduced by 89.1 %, and the average orientation angle error reduced by 96.8 %. In addition, the variances of the position error and the orientation angle error are also reduced by 97.94 % and 99.54 %, respectively.