Research On Motion Planning Control And Grabbing Strategy Of Autonomous Operation Robot

In industrial automation production lines,teaching industrial robots that complete handling,assembly and other operations need to have the function of grasping objects.However,this teaching operation robot cannot adapt to the modern production mode of multiple varieties and small batches.It is necessary to develop autonomous operation robots with perception and decision-making functions in order to realize the autonomous gripping function.In this paper,the universal robot operating system(ROS)is used as a development environment to study the realization of visual perception,path planning and motion control of autonomous robots;for the capture of multiple features of different objects,research on autonomous robots based on deep reinforcement learning Crawl strategy.The main research work and results of this article are as follows.1.Modeling and control system design of operating robot based on ROS.Based on the 6-degree-of-freedom operating robot and industrial CCD camera,a URDF model for robot modeling format in ROS is established,including:kinematics model,visualization model,dynamics model and collision model.The TF software library is used to realize the coordinate transformation between the parent and child joints of the robot in ROS,and the complete kinematic tree structure of the robot is constructed based on these information.Build the logic design and system cascade of functional modules such as image processing,motion planning,and servo drive:establish communication between ROS and robots and cameras through nodes to obtain robot status data and image streams;based on the ros_control software framework as joints The servo motor establishes a PID position controller to complete the planned path tracking of the robot.2.Path planning and motion control simulation experiment.Analysis and comparison of the rapid search random tree(RRT)method and the improved RRT_Connect algorithm.The average planning time of the improved RRT_Connect algorithm is 88.4%faster than the RRT algorithm.Based on OpenCV,the calibration plate is pose relative to the camera coordinate system and the hand-eye matrix are solved;through visual image processing,the pose of the target object in the robot is coordinate system is obtained;the RRT_Connect algorithm is used to realize the operation of the robot from the initial pose to the target position Posture path planning,through the servo control of the PID position controller to each joint,realize the tracking control of the robot end planning path;visual simulation of path planning and motion control in the Gazebo simulator to complete the autonomous gripping operation of the operating robot.3.Research on autonomous crawling strategies based on deep reinforcement learning.Aiming at the problem that the reinforcement learning is difficult to deal with the large state space and the long training time of the deep reinforcement learning(DQN)algorithm,the Actor-Critic algorithm combining Q-Learning and gradient strategy is used to study the robot crawling strategy.In view of the superiority of the pybullet simulator in the deep reinforcement learning training environment,the KUKA robotic arm in the simulator is used for the grab training simulation experiment.Aiming at the various characteristics of the gripping object,five randomly shaped parts were randomly generated as the gripping environment,the depth image of the Kinect camera was used as the input of the Actor network,and the actual action performed by the robot was used as the input of the Critic network.The end-to-end training model has been trained and tested to verify the effectiveness of the algorithm.Compared with the DQN algorithm in the same environment,the algorithm adopted in this paper has higher learning efficiency and crawl success rate,and fewer execution steps.