Research And Application Of Robot Intelligent Grasping Based On Deep Learning

In recent years,with the progress of flexible automation technology,the demand for intelligent grasping of robots in flexible production is increasing,while the traditional robot grasping technology has been unable to meet the increasingly complex automation requirements at this stage.In this paper,the vision sensor and robot are combined to design a robot grasping system based on deep learning detection algorithm.Finally,the improved YOLOv5 s detection algorithm is applied to the detection and grasping process of Dobot robot,which effectively improves the success rate of robot grasping.The main research contents and contributions are as follows:The robot grasping system platform is built.This paper analyzes and designs the scheme of robot grasping system based on the deep learning detection algorithm,analyzes the components of the system platform,expounds the relevant hardware selection and technical parameters,and completes the construction of the experimental platform.The calibration principle of robot grasping system is studied and the system calibration is completed.According to the imaging principle of the camera,the vision system with the eye out of the hand is established,and the conversion relationship between the pixel coordinate system and the world coordinate system is derived.The system calibration parameters are solved by Zhang Zhengyou calibration method and nine-point marking method.Research and improvement of object detection algorithm based on YOLOv5 s.The robot detection capture dataset is made and the number of label categories and label size distribution of the data set are analyzed.The YOLOv5 s object detection algorithm is used to achieve category recognition and position judgment of object objects.In order to improve the detection accuracy of the YOLOv5 s algorithm model,the following improvements are made in this paper: Firstly,the small target detection layer P2 is added to enhance the detection ability of the network for multi-scale features,so as to improve the detection effect of the model for small targets.Secondly,the weighted bidirectional feature pyramid network BIFPN is introduced and optimized to fuse more differentiated features through cross scale linking,enabling the algorithm model to achieve more efficient feature fusion capabilities and effectively solving the problem of insufficient feature fusion in the backbone network of robot grasping detection.Finally,the loss function is optimized to EIOU loss function to improve the accuracy of target location and the overall detection effect of the algorithm model.The experimental results show that the improved algorithm in this paper has a higher accuracy,with an average accuracy rate of 93.0%,which is 1.9%higher than the original YOLOv5 s model.The control principle of robot grasping system is studied.The D-H kinematic model of Dobot robot is established.The forward and inverse kinematics formula of Dobot robot is analyzed,and the correctness of the kinematics formula is verified by MATLAB simulation.Finally,a robot grasping experiment is designed to verify the effect of the grasping detection algorithm.The experimental results show that the success rate of robot grasping based on the improved YOLOv5 s detection algorithm is 94.4%,which is 2.4% higher than the success rate of robot grasping based on the original YOLOv5 s.It verifies the reliability of the robot detection and grasping system proposed in this paper,and also means that it can improve the intelligent degree of robot grasping and save the cost of human resources,which is conducive to the intelligent development of robot manufacturing.