Research On Robotic Time-Optimal Trajectory Planning And Robotic Force Control

Robot grinding,as an important field of robot application,has a great development over the past few decades.In most applications,robots need to be taught manually to obtain the grinding paths,which is quite laborious and unable to meet the needs of intelligent manufacturing.Nowadays,with the continuous expansion of the application range of robot,the gradual maturity of sensor technology and the development of intelligent algorithms,intelligent manufacturing is becoming possible.And how to make the robot grinding process automatically and efficiently is the key point to realize intelligent manufacturing.This article focuses on how to improve the efficiency of robot grinding and how to realize the robot's perception of the semi-finished product surface to achieve automatic grinding.The main content is divided into the following three parts:(1)To improve the efficiency of robot grinding,the robot's time-optimal trajectory planning method is studied to shorten the running time of auxiliary strokes.A numerical integration-like(NI-like)time-optimal trajectory planning approach combined with the iterative learning algorithm is proposed.In the proposed approach,the robot path is expressed by a scalar path coordinate and discretized into N points.The motion between two neighbouring points is assumed to be uniformly accelerated motion,so the time-optimal trajectory that satisfies constraints is obtained by using equations of uniformly accelerated motion instead of numerical integration.To improve dynamic model accuracy,the Coulomb and viscous friction are taken into account(while most publications neglect these effects).Furthermore,an iterative learning algorithm(ILC)is designed to correct model-plant mismatch by adding an iterative compensation item into the dynamic model at each discrete point before trajectory planning.(2)To avoid the model-plant mismatch phenomenon during time-optimal trajectory planning,a dynamic model-free reinforcement learning approach is proposed.In the proposed approach,we use an improved SARSA(State-action-return-state-action)algorithm and a two-step method for finding the time-optimal trajectory and ensuring the feasibility:Firstly,using improved SARSA algorithm to find a safe trajectory that satisfies the kinematic constraints through the interaction between the reinforcement learning agent and kinematic model.Secondly,using improved SARSA to find the optimal trajectory through the interaction between the agent and the real world,and assure the actually measured torques satisfy the given limits at the last interaction.(3)To realize the robot's perception of the semi-finished product surface to make automated grinding,the robotic force control method based on an semi-finished product surface is studied.A hybrid force/position control scheme based on adaptive iterative learning algorithm is proposed.The control method is composed with two steps:1).An iterative learning control law was designed based on the impedance model of robot-environment in dynamic interaction task.This control law coped with the unknown parameters and disturbances by adding the iterative term to the proportional-derivative(PD)feedback structure.Meanwhile,a Lyapunov energy function was designed to prove the convergence of the control law.2).The adaptive iterative learning control law was then combined with the force/position hybrid control method to design the Constant-force curved-surface-tracking control scheme with robotic manipulator.By using this method,the initial path of the robot for grinding semi-finished product surface can be obtained automatically.