Design And Implementation Of Human Motion Intelligent Control And Simulation System

Human motion simulation plays an important role in the development of humanoid robots,movie animation,and virtual reality.Simulation actions are generally implemented based on the method of capturing data.However,this method is usually not interactive with the environment.In order to improve the authenticity and interactive ability of simulation movements,this paper adopts methods of reinforcement learning for training,and designs and implements a simulation system with functions such as the establishment of robots.the control of motion models,and the verification and evaluation of simulation actions.The specific work of this paper is as follows.First,we complete the work of human motion data processing and conversion.A data conversion algorithm is constructed to convert the joint data expressed in Euler angles to quaternion form,which avoids the problem of universal lock and is convenient to use.Secondly,a virtual robot RLM(Reinforcement Learning Model)is constructed based on the Bullet physics engine.Its proportions,joint degrees of freedom,and mass are similar to the human body.The PD controller method is used to calculate the joint torque,and the controller parameters are optimized offline to generate the optimal torque to control the joint.Thirdly,we propose a motion control method based on reinforcement learning.The method uses the collected motion data as the target value,and designs a multi-reward function to train the RLM,and finally outputs a suitable torque to drive the model to complete the action.The results show that using this method can have a higher reward return value.verifying the effectiveness of the method.Finally,an evaluation method based on similarity is proposed to verify and evaluate the simulation action.This paper designs two environments for applying force to RLM and throwing cubes.Extract the joint angle and reward value of RLM,and experimentally verify that the reward value curve of RLM first decreases and then increases under the environment of external interference,and finally resumes action.At the same time,the simulation motion senerated by RLM coincides with the joint angular motion trajectory of reference target data.