Research On Combined Control Method Of Somatosensory And Depth Reinforcement Learning For Cooperative Manipulator

In recent years,as the concept of human-computer collaboration has received more and more attention,collaborative robot technology has developed rapidly,and its stability and dexterity have replaced humans in fulfilling a large number of repetitive or dangerous tasks.At the same time,as people's exploration of the world has increased,there have been a lot of harsh environments that are not suitable for human beings to go to work.Based on this background,this paper mainly studies a robotic arm teleoperation control system based on the somatosensory method,which enables the operator to control the smart robotic arm to complete complex tasks efficiently and naturally by using the body movements and gestures from the far end.The main research contents of the thesis include:(1)The body motion capture device Kinect is used to acquire the human body motion position,and the motion mapping algorithm is completed,so that the human body motion is more accurately mapped to the robot arm,and the natural body sense control of the robot arm is experimentally performed.Through the improved Socket communication,the long-distance reliable transmission of data is completed,and the working environment is observed through the visual feedback system,thereby realizing the master-slave teleoperation task of the robot arm.(2)The kinematics of the manipulator was analyzed and the manipulator was modeled using the D-H method.A method based on the skeletal vector solution angle is proposed.The joint position data obtained by Kinect is used to calculate the angle between the bones.A filter suitable for this experiment is designed for the smooth processing of bone data.(3)In order to solve the problem of the adjustment of the end position adjustment in the somatosensory control,this paper introduces the deep reinforcement learning algorithm and builds the simulation environment for large-scale training.Finally,the somatosensory and deep reinforcement learning combination control method is used to complete the higher precision position adjustment for the end of the mechanical arm.(4)In order to better observe the working environment of the slave arm,this paper introduces a virtual reality technology(VR)with high authenticity and high immersion into the visual feedback system.The communication protocol is used to accept the working environment video stream information,and the Unity 3d software is used to build the virtual reality environment to realize the real-time rendering display of the video stream in the virtual reality glasses,thereby completing the construction of the stereoscopic feedback system,so that the operator can Observe the scene of environment from the first person Perspective.(5)This experiment was done in both Windows and Ubuntu systems.Through the UDP protocol,the cross-system transmission of bone data is realized,and the basic somatosensory control of the robot arm is realized by an algorithm.The Unity and ML-Agents machine learning agents are used to build an end-position autonomous adjustment training environment based on deep reinforcement learning.Finally,this paper uses the Baxter dual-arm robot to build a real robotic arm experimental platform.The final experimental results verify the effectiveness of the control method.