Research On Uncalibrated Visual Servo Technology For Hyper-Redundant Manipulators

Visual servoing introduces the visual feedback information into the control system of the robot,which significantly enhances its ability of autonomous working and perceiving the environment.Compared with the traditional calibrated visual servo,the uncalibrated visual servo technology does not need to accurately calibrate the camera parameters and the relative pose relationship between the manipulator and the camera,which further improves the robustness and application range of the visual servo system.Meanwhile,due to the advantages of hyper-redundant manipulator in flexibility and fault-tolerance,it plays an increasingly important role in the complex and dangerous industrial and aerospace fields,and it has also become a hotspot in the field of robot.Therefore,it is of great significance to apply uncalibrated visual servo technology to hyper-redundant manipulators.In actual work,complex environmental factors will reduce the success rate and accuracy of the system to obtain visual feedback information,thereby further affecting the accuracy and reliability of the visual servo system.Complex structure and parameters of the hyper-redundant manipulator will affect the performance of system as well.According to the above problems,this paper will conduct research on the uncalibrated visual servo technology of hyper-redundant manipulators in the following aspects:Aiming at the acquisition of visual feedback information in uncalibrated visual servo system,a cooperative target with its identification and localization method are designed.First,a design scheme of a cooperative target with 9 feature points is proposed.It distinguishs each feature point based on the affine invariance and geometric constraints,and show certain robustness to the situation of partial feature points missing.Then,adaptive threshold and local optimum strategy are adopted for edge extraction.According to the coding information and geometric constraints of the feature points,an identification method based on ideal edge fitting is proposed,which is accurate in various scenarios.Finally,the improved non-maximum suppression method is used to further screen the edges of feature points,and a method which combines the secondorder sub-pixel edge localization method based on local area effect and the least squares ellipse fitting method is used to locate the feature points.According to the characteristics of the hyper-redundant manipulator,an uncalibrated visual servo system based on projective homography is designed.First,an improved homography-based task function is proposed,which is robust to the missing of partial feature points and further reduces the dimensionality.Subsequently,in order to make the system robust to the kinematic parameter errors of complex hyperredundant manipulators,a neural-network-assisted robust filtering scheme is designed to do online estimation of the Jacobian matrix which directly maps the task function and the manipulator joints.In the robust filtering scheme,the neural network plays the role as a corrector which compensate the estimation error of Jacobian caused by the positioning error of feature points.In addition,in view of the problem that the estimated Jacobian matrix is ill-conditioned due to the large degree of freedom of the hyperredundant manipulator,the singular value filtering algorithm is introduced to limit its minimum singular value.Based on the estimated Jacobian matrix,controllers for static positioning tasks and dynamic tracking tasks are constructed respectively.The superiority of the system is verified by the simulation and experimental results.The optimization of the uncalibrated visual servo system based on projective homography is completed.In order to solve the problem that the joint angle may exceed the limit in the process of visual servo,the gradient projection method is used to plan the self-motion of the manipulator joints based on the redundancy of the hyperredundant manipulator joint space.It avoids the damage to the mechanical structure caused by the overrun of joint.Aiming at the limitation of the fixed servo gain in static positioning tasks,an adaptive adjustment method of servo gain based on Q-Learning algorithm is proposed.It selects the optimal servo gain from the action set in real time according to the Q-Table and the current error of task function.The results of simulation and experiment show that,compared with the traditional fixed servo gain,adaptive servo gain can speed up the convergence of system while ensuring the stability of the system.The uncalibrated visual servo method based on robot dynamics is researched.A system with eye-to-hand configuration is designed to alleviate the difficulties caused by the complex dynamic model of a hyper-redundant manipulator.The method for solving the forward and inverse dynamics problem of the hyper-redundant manipulator is introduced,and the kinematics of the feature points and the hand-eye mapping relationship of the system are modeled.Subsequently,a controller based on a depthindependent Jacobian matrix is used to make the unknown internal and external parameters of the camera appear in the closed-loop dynamics equation of the system in a linear form,and the unknown parameters are estimated by an adaptive method.Finally,the stability of the uncalibrated visual servo system is analyzed and proved based on the Lyapunov theory.The simulation results verify that the system can achieve accurate positioning of static targets.