Data-Driven Industrial Robot Self-Learning Trajectory Planning Method

Trajectory planning is an important prerequisite and guarantee for high-performance robot operation.The continuous trajectory requires high accuracy,existing continuous trajectory planning algorithm has a large amount of calculation,trajectory planning algorithm for continuous trajectory still needs to be researched and developed.Existing optimal trajectory planning algorithms rely on dynamics modeling,the process is tedious,complicated and computationally intensive.The PID controller is the most commonly used controller in the industry.Due to the non-linear characteristics of robot dynamics and uncertain interference,it is very difficult and tedious to design and adjust PID controller parameters to achieve high-precision tracking control.Aiming at these problems,this paper assumes that the dynamic model is unknown and the trajectory planning algorithm is optimized and theoretically analyzed and simulated based on the PID controller.Experimental research is carried out.Aiming at the requirements of robot operation index,a trajectory planning scheme with intelligent learning ability is proposed.By analyzing the difficulties and problems encountered by industrial robots in actual machining,the overall scheme of intelligent trajectory planning algorithm is designed,including robot basic trajectory planning module,offline learning module and online learning module.The basic trajectory planning module obtains the forward and inverse solution relationship of the robot by establishing the kinematics model,and processes the geometric trajectory through spline interpolation.By establishing a mathematical model of the small line segment and using the forwardlooking algorithm to do speed plan of the continuous trajectory,finally a stable trajectory is obtained by interpolation.Offline learning for tracking error of robot non-repetitive trajectory,by designing neural network structure and input feature selection,using the powerful model fitting and generalization ability of neural network,using robot operation data training neural network to learn the map between ideal trajectory and corresponding reference trajectory.Given each ideal state trajectory as a neural network model input,the trained neural network model generates a new reference trajectory as the output state,and the robot system achieves the purpose of reducing the tracking error by following the new reference trajectory.Online learning focus on the tracking error and disturbance problem of the robot's repeated trajectory.The online learning framework is designed based on it-erative learning and radial basis function neural network,including the design of iterative learning framework,the design of neural network structure,the design of neural network learning law,proof of the stability of online learning.According to the actual error of each operation of the robot's repeated trajectory,the online learning algorithm updates the neural network parameters to generate a new reference trajectory for the robot to run until the error reduces below the threshold.The simulation and experimental verification show that the basic trajectory planning algorithm can meet the performance requirements under the premise of small calculations;the offline learning algorithm can effectively reduce the tracking error of non-repetitive trajectories at around 60%;through continuous iterative learning,the tracking error of the repeated trajectory is reduced at around 80 %.