Research on Intelligent Grasping System for General Objects Based on Deep Learning

Aiming at the traditional mechanical arm grasping limited to fixed poses, specific shapes of objects, and non-real-time recognition, general object recognition and intelligent grasping system based on deep learning is designed. The system applies the YOLO-V4 object detection model and uses the Intel RealSense D435i depth camera to collect color images of different postures and positions of the general target object, which is used to train the target detection model, and proposes an adaptive labeling best capture the method of location improves the accuracy of crawling. Then, according to the hand-eye calibration result of the robot grasping system and the homogeneous transformation matrix of each joint of the robotic arm, the three-dimensional coordinates of the best grasping point in the image coordinate system are returned; the system then performs inverse kinematics to solve the obtained coordinate information to determine the pending Grasping the position of the object, and finally using the proposed improved RRT-Connect multi-stage motion planning optimization algorithm for motion planning, after obtaining the planned path, the robot arm is guided to efficiently grasp the target object.