Research And Application Of Vision-based Manipulator Target Detection And Obstacle Avoidance Path Planning

As one of the basic equipment for the development of automation technology,the manipulator has played a pivotal role in some industries after long-term development.However,in today’s highly intelligent era,the manipulator still lacks the ability of active perception and safe real-time interaction.In this paper,the binocular camera is used as the visual sensor of the manipulator,and the depth learning technology is used to realize automatic visual detection on the 6-DOF manipulator platform,and the path planning algorithm in the operation of the manipulator is explored to improve the autonomy and intelligence of the manipulator.Firstly,the manipulator system based on ROS is built.The kinematics of the manipulator is analyzed.The manipulator model is configured by Move It! in the ROS system.The hand-eye calibration experiment is carried out by using the Ar Uco label,and the transformation matrix between the manipulator and the visual system is obtained.The octree map of the manipulator platform in the workspace is obtained by using the point cloud information.Secondly,for the task of determining the target object and its pose of the manipulator system,the idea of determining the category and range by target detection,and then determining the grasping pose by grasping detection is designed.Aiming at the problems of complex network structure and large amount of calculation in the YOLOv4 target detection algorithm,the lightweight design of YOLOv4 algorithm is carried out.The lightweight network Mobile Net V3 is introduced to replace the CSPDarknet53 as the backbone feature extraction network,and the network is further optimized by depthwise separable convolution.The lightweight Small-Mobile Net V3-YOLOv4 network is obtained,and the improved network is easier to be deployed on the embedded platform.The network model before and after the improvement is trained and evaluated on the selfmade dataset of eight classes of graspable objects,and it is verified that the improved model can significantly improve the detection speed.Then,the grabbing attention is focused on the target area by using the results of target detection.Combined with the grabbing detection algorithm based on CNN,the grabbing pose is obtained.Then,the obstacle avoidance path planning of manipulator is studied.The RRT,RRT-Connect and RRT* algorithms are analyzed and compared.The RRT* algorithm is improved by adding target-oriented node strategy and adaptive probability strategy,which reduces the running time and avoids falling into local minima.The three-dimensional simulation experiment of obstacle avoidance path planning is carried out.The path generated by planning is optimized by cubic polynomial method.Path planning is carried out on the manipulator simulation platform,and the executable continuous smooth path trajectory of the manipulator is generated.Finally,the verification experiment is carried out on the manipulator platform.The target detection model is deployed to the manipulator system to test the application effect of the algorithm.The experiment of identifying and grasping the target object is completed by using the manipulator system,which shows the reliability of the target object detection algorithm and the obstacle avoidance path planning algorithm proposed in this paper.