The Research On Vision Based Robotic Arm Grasping Target Detection And Motion Planning

Nowadays,intelligent robots are used to provide services for human labour and production and daily life,which is one of the core functions of intelligent robots.Currently,most of the robotic arm grasping tasks are aimed at the grasping of known objects,but in the face of unknown objects of different shapes and postures in daily life service scenarios,there is an urgent need to find new solutions.In order to study the intelligent recognition method and grasping operation of unknown objects an d further improve the intelligence of the robotic arm,this paper uses the UR5 robotic arm,D435 i depth camera and AG-95 two-finger gripper as hardware and the grasping target detection algorithm and path planning algorithm as software to realise the intelligent recognition and grasping of unknown objects in daily life scenarios and completes the simulation in the ROS(Robot Operate The software is used to realize the intelligent recognition and grasping of unknown objects in daily life.The main research contents of this paper include:(1)Analyse the process of grasping tasks in daily life scenarios,determine the hardware and software required to realise the recognition and grasping tasks;and analyse the camera calibration principle,complete the camera calibration experiments and obtain the cameras’ internal reference matrix;determine the "eye in hand" handeye solution,analyse its basic principle and conduct hand-eye calibration experiments to obtain the relative coordinate transformation matrix between the camera and the robotic arm.To determine the "eye in hand" hand-eye solution,analyse its basic principle and conduct hand-eye calibration experiments to obtain the relative coordinate transformation matrix between the camera and the robot arm,so as to lay the foundation for the grasping task.(2)Designing a deep learning-based grasping target detection network.The YOLOv4 target detection algorithm network was improved by replacing the rotating anchor frame with a line segment type of grasping position information,and simplifying the model structure to lighten the network and reduce the computational effort.When RGB images are input to the network,the line segment type grasping parameters of the object and the object category are output;the self-made dataset is made and trained to obtain the grasping target detection network model and tested,and the result is that its detection speed is 30 frames/second,and the success rate of object recognition prediction is 84%,achieving intelligent recognition of unknown objects.(3)Planning of robotic arm motion process.Modeling the DH method for the UR5 robotic arm and performing kinematic solution analysis of forward and inverse kinematics,as well as verifying in MATLAB to prove the correctness of the solution;comparing the advantages and disadvantages of RRT and RRT* path planning methods,adding bias strategy to the RRT* method for improvement,and obtaining a 3D spatial planning speed of 0.023 seconds;smoothing the robotic arm end trajectory with three times B spline curve Finally,the path planning algorithm is imported into Move It for motion planning,and the running trajectory can be obtained.(4)Under the ROS system,the gripping target detection algorithm and the path planning algorithm are integrated and configured to build a simulated gripping experiment platform and complete the simulated gripping experiment of the robot arm.The results show that the average grasping success rate in the simulation grasping experiment is 81.67%,which verifies the feasibility of the grasping task in daily life scenarios.