Manual Dragging Teaching And Collision Detection Based On Robot Dynamics Model

As an important direction of robot intelligence,human-robot collaboration has received more and more attention.Compared to collaborative robots,industrial robots are cheaper and not equipped with joint sensors or multi-dimensional sensors,which will encounter greater difficulties to implemente human-machine collaboration.At present,considering the cost,the main ways to achieve human-machine collaboration of industrial robots are based on robot dynamics.In this context,this paper has made related research on drag-teaching and collision detection technology based on robot dynamics model.In this paper,the multi-axis serial industrial robot is taken as the research object,and the dynamic models of the robot are constructed,which includes the processes of dynamic equation construction and parameter identification.In terms of basic dynamics,the inverse dynamics equations of the robot are established and linearized based on the Newton-Euler method.Then,the multilevel Fourier series trajectory is used as the excitation trajectory.Finally,the weighted least square method is used to obtain the basic dynamics values.Next,the non-linear friction characteristics inside the robot joint are further studied and elastoplastic friction model is used to model the joint friction torque of the robot.Elastoplastic model uses an elastic friction unit and a spring with stiffness to effectively describe the pre-sliding friction effect and stribeck effect.In the parameter identification of the elastic friction model,the particle swarm optimization algorithm is adopted,and the error between the calculated torque and the measured torque is used as the objective function to establish an iterative optimization scheme.For the wave friction effect existing in the joint reducer,a method of combining the Fourier series and BP neural network is designed to model the wave friction torque,which improve the accuracy of the dynamic model of the robot.Then,based on the robot dynamics and position control,this paper proposes the sensorless robotic drag-teaching control methods,including drag-teaching based on joint space,drag-teaching based on Cartesian space,and drag-teaching control in motion.Based on the robot's dynamics model and friction model,the generalized momentum estimation method is used to detect the external torque and force,and the admittance control method is used to implement drag-teaching.In order to offset the influence of the static friction of the joint,the start stage of the drag-teaching is improved to improve the smoothness of the drag.Then,based on the existing dynamic model,this paper analyzes the variation characteristics of the torque error when the robot is under collision.Collisions are classified into hard collisions and soft collisions based on the existence of vibration phenomena,and a collision detectionalgorithm based on the envelope-like curve is proposed.Finally,this article takes industrial six-axis robots as the experimental objects,and conducts experiments from three parts: robot dynamics model,drag-teaching,and collision detection.Firstly,the dynamic model of the robot is identified experimentally,which includes the basic dynamic model,the elastic friction model and the wave friction model,and a complete set of robot dynamic model is obtained.Based on this model,the robot's drag-teaching and collision detection methods are verified.The drag-teaching experiment part carried out three experiments: drag-teaching in joint space,drag-teaching in Cartesian space,and drag-teaching during motion,which verify the effectiveness and smoothness of dragging.The envelope-like collision algorithm was tested under conditions of hard collision,soft collision,and human-machine collision,which verify the sensitivity and safety of the collision detection algorithm.