Inverse Dynamics Modelling Of A Multi-degree-of-freedom Robotic Arm Based On A Self-attentive Mechanism

The industrial robotic arm is gradually developing towards the direction of intelligent,integrated and high-performance control,of which robotic arm control is one of the most important parts,and the robotic arm inverse dynamics model,as an important element of the robotic arm control feedforward term,is closely related to the robotic arm control accuracy.Therefore,it is very important to establish a high precision inverse dynamics model of the manipulator.In this context,an improved model based on the self-attention mechanism is proposed to complete the inverse dynamics modeling of the manipulator based on the deep learning method based on the UR5 manipulator,and the prediction accuracy and efficiency of the proposed improved model are verified by comparing it with the current common inverse dynamics network model used for the manipulator.The main work content of this paper is as follows:Firstly,the kinematics analysis and simulation of the manipulator are carried out.Before the analysis,the coordinate system about the manipulator was established by DH parameter,and the homogeneous transformation matrix of the manipulator was deduced by rigid motion.The forward kinematics analysis and velocity kinematics analysis of the manipulator were completed by deducing the pose of the end-effector of the manipulator through the homogeneous transformation matrix given the joint Angle.According to D-H parameters,the manipulator model was established through the robot toolbox in matlab,and the forward kinematics simulation calculation and Jacobian matrix calculation were completed by using the built-in function of the toolbox.The results showed that the simulation calculated values and the theoretical calculated values were within the error range,which verified the correctness of the forward kinematics analysis and velocity kinematics analysis.The working space of the manipulator is described by Monte Carlo method.Secondly,the dynamic equation of the manipulator is established based on Newton-Euler method and the simulation software is used to complete the simulation calculation.According to the results of kinematic analysis of the manipulator,the velocity and acceleration formulas of the connecting rod and the center of mass were derived,and the inertia tensor was derived.The dynamics equation of the manipulator was established according to the Newton-Euler recursive formula,friction model and flexible joint model,and the calculation of the dynamics equation was completed.The dynamic simulation software Adams was used to establish a virtual prototype,set parameters and add elastic torsion springs to the joint to simulate the joint flexibility to complete the simulation calculation.It was found that the established coupling model has significantly improved the accuracy compared with the rigid body model.Thirdly,the inverse dynamics modeling of the manipulator with deep learning method is completed.Simple neural networks and cyclic neural networks and their variants are summarized by applying deep learning methods to robot dynamics problems.Self-attention mechanism and Transformer model based on self-attention mechanism are introduced.An improved model based on self-attention mechanism is proposed and related code is written.Finally,the experimental platform was built and the results were analyzed.This paper introduces the structure of UR5 manipulator,and expounds the hardware structure and joint characteristics of the manipulator used in research.The robot arm model was constructed and processed by Creo,a 3D modeling software.The control and data acquisition of the robot arm were completed.Communication was established with the actual robot arm through PC,and the robot arm was controlled based on ROS2 system to complete the collection and processing of the required data.Compared with the current common network models,it is verified that the proposed improved model has obvious improvement in prediction accuracy and efficiency.