并联机器人视觉盲区末端位姿检测方法

Pose detection for visual blindness of parallel robot

摘要:

为解决并联机器人末端执行器受机构支路遮挡造成的双目视觉盲区末端位姿错误检测问题,提出一种运动学正解结合混合优化RBF神经网络(RBFNN)误差补偿的视觉盲区末端位姿检测方法。首先在非视觉盲区采集RBFNN训练样本,其中运动学正解为输入样本,运动学正解和视觉检测位姿的差值为输出样本；然后进行训练,并采用GWO(Grey Wolf Optimization)算法和LM(Levenberg-Marquardt)算法混合优化权值；最后将训练好的网络用于视觉盲区,通过对运动学正解进行误差补偿以提高末端位姿检测精度。实验结果表明,与未补偿的检测方法相比,混合优化RBFNN补偿后的末端位姿检测方法,其末端位姿分量x,y,z,γ的误差平均绝对值分别降低了54.4%、67.7%、54.7%和52.9%,误差标准差分别降低了52.9%、62.8%、51.9%和58.8%,验证了所提方法的有效性。

Abstract:

In the pose detection for a parallel robot based on binocular vision, error detection can be caused by end-effector being obscured by the branch of the mechanism. To solve the problem, a pose detection method for visual blindness based on the direct kinematics compensated by a hybrid optimization RBF neural network (RBFNN) is proposed. Firstly, RBFNN training samples are collected in non-visual blindness, where the direct kinematics is the input sample, and the difference value between direct kinematics and pose detected by binocular vision is the output sample. Then, Grey Wolf Optimization (GWO) algorithm and Levenberg-Marquardt (LM) algorithm optimize the weights in the training process. Finally, the hybrid optimized RBFNN having been trained is applied to compensate the error of direct kinematics to improve the accuracy of pose detection for visual blindness. Experimental results show that compared with the uncompensated pose detection method, when the pose detection method compensated by a hybrid optimization RBFNN is applied, the mean absolute value of error for pose component x, pose component y, pose component z and pose component γ are reduced by 54.4%, 67.7%, 54.7% and 52.9%, respectively; the standard deviation of error for pose component x, pose component y, pose component z and pose component γ are reduced by 52.9%, 62.8%, 51.9% and 58.8%, respectively. The results verify the effectiveness of the proposed method.