Hybrid Optimization Strategy For Redundant Robots Based On Neural Dynamics

Redundant robots have great application potential because of their superior flexibility and excellent accuracy.However,the design of existing motion planning algorithms and solvers limit their further popularization to a large extent.Most of the traditional motion planning algorithms for redundant robots consider simple tracking tasks in structured scenarios.This can simplify the algorithm design steps and reduce the computing power requirements of the computer,but it is difficult to give full play to the flexibility of the redundant robot,and cannot meet the performance and functional requirements of the robot in complex environments and tasks.The consideration of multi-index problems is also mostly based on pseudo inverse matrix method,which is not intuitive and flexible in the design of optimization problems,and has limitations in the speed and accuracy of solution.Therefore,there is a large space for research in the design of the motion planning algorithm and solver of redundant robots.For the motion planning problem of redundant robot,the hybrid optimization design with multiple optimization criteria or functional criteria can give full play to its redundancy characteristics.The performance optimization brought by the optimization criteria and the additional functions endowed by the functional criteria can effectively improve the adaptability of the robot to the complex environment and task,and also make the robot system more intelligent.With the expansion of robot applications,its working environment will become increasingly complex,and its task requirements will become more and more diverse.The design scheme of hybrid optimization with multi criteria for robot motion planning will play an important role.In addition,in order to ensure the control performance of the robot and cope with the problem of increasing solving complexity caused by complex criteria,it is necessary to design a robust and efficient solver.It is an effective scheme to design the motion planning solver of redundant robot based on neural dynamics by using the parallel characteristic of neural network.In order to improve its performance in the process of robot motion planning and strengthen its execution ability;aiming at key problems,including the problem about end-effector tracking ability of redundant robots,the obstacles collision problem,the joint drift problem in repetitive movement and the noise problem;taking to improve the motion planning performance of redundant robot system as the goal;taking the Kinova Jaco2 manipulator and the mobile robot independently designed by the laboratory as objects,the design,analysis and application of the hybrid optimization strategy for redundant robots based on neural dynamics are carried out from the design of kinematic equality constraints,functional constraints,objective function and solver in the motion planning of redundant robot.The main work of this thesis is summarized as follows.1)A novel kinematic equality constraint is designed to overcome the end-tracking performance defects of traditional robot motion planning schemes.Through theoretical analysis,it is found that the robot motion planning scheme based on a dual neural network has the defects of end-effector position error accumulation and limited initial position configuration.Based on the analysis of the traditional scheme control law,a novel equality constraint is designed which can eliminate the initial error and overcome the problem of error accumulation during the task execution.The effectiveness of the equation constraint in overcoming the above defects and the accuracy of robot control are verified by simulation and experiment.2)A novel hybrid functional constraint is designed to enable the robot to simultaneously avoid obstacles,maintain end-effector posture and avoid joint limit in the process of motion planning.The end-effector posture maintaining method does not require precise robot attitude data and can keep the robot in any reachable end-effector posture.Simulation and experimental results on the fixed manipulator and the mobile robot verify the effectiveness of the hybrid functional constraint for robot motion planning.3)Two hybrid objective functions are designed to achieve unbiased repetitive motion of the robot and maximize its manipulability.The two hybrid objective functions are designed by using the traditional fixed weight combine method and the proposed time-varying parameter combine method respectively.The time-varying parameter combine method improves the motion planning performance of the robot by redistributing the flexibility of the redundant robot in time.Through the robot simulation of the two hybrid objective functions on the repetitive motion tasks,the effectiveness of these two functions to simultaneously consider the joint drift problem and the manipulability maximization problem in the repetitive motion of robots;the result comparison verifies the characteristic of the two combine method and the superiority of the time-varying parameter combine method.4)An anti-noise solver is designed to suppress the noise in robot motion planning,especially the periodic noise that may be generated when the robot executing repetitive motion.On the basis of proving the global stability of the solver,the zero stability and uniform convergence of Euler-type discrete solver and Taylor-type discrete solver are verified.Through robot simulation and experiment,the proposed solvers have excellent anti-noise ability and accurate solving ability under different activation functions and different periodic noises.5)A hybrid optimization scheme was designed to solve the joint drift and torque norm minimization problems in the process of repetitive motion of robot at the acceleration layer.It is applied to a robot to verify the effectiveness of the hybrid optimization scheme.