基于图像的非标定视觉反馈控制机器人全局定位方法

摘要

To address the robotic uncalibration global positioning problem,we studied a Kalman filtering (KF)unite feedback Elman neural network (ENN)for learning nonlinear mapping between robot image-space and movement-space,then propose an image-based visual feedback control method.First,suboptimum states were obtained by ENN global learning for robot global positioning to build the system state equation and observation equation,and further use KF to estimate the image Jacobin matrix.Second,KF also fine-tuning the ENN's weights in real time,the KF cooperative working with ENN not only meets the global stability of the robot global positioning,but also exhibits a certain adaptability to the dynamic environment.Finally,under conditions of uncalibrate the camera pa-rameters,many positioning comparison experiments had been carried out with six degrees of freedom"eye-in-hand"robotic to verify the effectiveness of the proposed uncalibration image visual serving method.%针对机器人非标定全局定位问题,研究Kalman滤波(Kalman filtering,KF)算法联合反馈型Elman神经网络(Elman neural network,ENN)学习机器人图像空间与运动空间非线性映射关系,从而建立基于图像的视觉反馈控制方法.首先利用ENN学习得到机器人全局定位的次优状态,以此为系统状态向量构建伺服系统状态方程与观测方程,进而利用KF估计得到机器人图像雅可比矩阵.其次,采用KF对ENN网络权重进行在线微调,KF联合ENN满足机器人全局定位稳定收敛的要求,并对环境干扰具有一定的自适应性.最后在摄像机参数未标定条件下,进行六自由度机器人"眼在手"(eye-in-hand)定位比较试验,结果验证了提出的非标定视觉伺服控制方法的有效性.