Research On Robotics Uncalibrated Hand-Eye Visual Servo System

In the recent years,the applications of robot vision have been gradually widespread in various fields.The complex design and calibration process is involved in the traditional robot vision system,such as camera calibration,hand-eye calibration and robot coordinate system calibration.Therefore,the non-calibrated technique has received extensive attention in these years.In this dissertation,some key techniques are discussed for the uncalibrated visual servo(UVS)system of the manipulator.For multi-degree-of-freedom manipulator,the kinematics analysis of the manipulator is first performed,and various Jacobian matrices of the manipulator visual servo system are established.Combined with the visual model of the robot,the robotic visual servo control strategy is introduced to lay a theoretical foundation for the UVS system.However,the parameters of the noise characteristics in the actual control system are unknown,which limits the application of the classical Kalman filtering algorithm.Consequently,in this paper,the classical Kalman Filter(KF)is analyzed firstly,subsequently a Kalman gain adaptive method based on regression estimation of noise parameters under the condition of unknown noise parameters is summarized and derived as an online estimation algorithm.The Projective Homography Based uncalibrated(PHUVS)and the traditional Image Based uncalibrated Visual Servo(IBUVS)are simulated under different tracking tasks.Finally,it is concluded that when adaptive Kalman filter(AKF)combined with PHUVS control strategy in uncalibrated system,the response time and image steady-state error,position steady-state error and pose steady-state error is close to the system after calibration,and the performance is superior to IBUVS.According to the superiority of particle filter(PF)in state estimation of nonlinear systems,combined with Kullback-Leibler Divergence(KLD)algorithm,the number of particles is adaptively adjusted,and the adaptive particle filtering(APF)is used as an online estimation algorithm in UVS systems.Under the condition of common Gaussian noise and Gamma noise added in UVS system,and the performance of uncalibrated visual positioning is compared while KF,AKF,PF and APF are used as online estimation algorithms.Through experiments,it is concluded that under the assumption that the system noise is Gaussian,AKF has better system performance index,and APF has better performance under non-Gaussian noise conditions.Finally,the basic principle of optical flow detection is analyzed.Combining the relationship between optical flow estimation and camera self-motion,the convolutional neural network VSFlowNet is used to estimate the camera self-motion information while using AKF to estimate the hand-eye Jacobian matrix.The response time and control accuracy of the UVS system are verified using Gazebo associated with Moveit!.The experiment was carried out in two experimental environments without adding disturbance and adding disturbance,and good system performance of which are obtained.