Research On Object Detection And Grabbing System Of Robot Based On Deep Learning

Robot grasping technology is widely used in many fields such as industry and daily life.In order to meet the needs of social development,robots need stronger environmental adaptability.with the recent research hotspots of artificial intelligence,the deep learning target detection technology can be combined with robot grasping to improve the intelligent ability of robots and complete object detection and automatic grasping tasks in complex industrial scenes.In this paper,deep learning,image processing and robotic arm grasping technologies are integrated,and using the depth camera sensor to achieve the robot perceive ability to the environment.Then,a powerful robot target detection and grasping system experiment platform is built.The main research work is as follows:First of all,according to the task requirements of the robot’s target detection and grasping system,building the whole architecture system,constructing an eyes-in-hand system.Using the Zhang Zhengyou algorithm to accurately convert the information between coordinate systems.Secondly,in the target detection part,aiming at the problem of false detection and missed detection of small target objects in the widely used YOLO V3(You Only Look Once)algorithm,an improved multi-scale feature fusion algorithm is proposed,which increases the width of the feature extraction of the backbone network.The pyramid structure is used to further deepen the depth of the convolutional network,and the residual unit structure is also added to solve the gradient disappearance and gradient explosion problems,then the detection accuracy of multi-scale targets is improved.With the depth point cloud image,PCA(Principal Component Analysis)technology is used to judge the main direction of the target.Thirdly,in the part of the robotic arm,a six-axis serial robot,which is independently developed in the laboratory,according to the D-H parameter table of the new robot,the forward and inverse kinematics equations are deduced.Based on the artificial potential field method,the robot grasping path planning RRT(Rapidly-Exploring Random Trees)algorithm,which improves the efficiency of robot grasping,optimizes the motion planning for placing target objects,and provides a basis for robot motion planning.Finally,on top of the above work,the robot object detection and grasping experiments are carried out.A comparative experiment is carried out on the target detection algorithm to verify the effectiveness of the proposed algorithm;the ROS robot operating system and visual sensor work together to get feedback information,using the Move It motion planning plug-in to complete the grasping and releasing object tasks with the end gripper of the robotic arm.The experimental results show that the proposed robot target detection and grasping system based on deep learning has practicality and promotion value.