Intelligent Grasping Method of Multi-joint Manipulator Based on Deep Learning

Mechanical arm is widely used in various fields of industrial production, and grasping operation is the most common application of mechanical arm in assembly line operation. When grasping an object, the robot arm should first acquire the position of the target object and adjust its posture according to the position of the target object to complete the grasping task. At present, the construction of hand-eye system based on depth camera and traditional mechanical arm, which enables the mechanical arm to grasp intelligently and autonomously, is the mainstream development trend of mechanical arm. The visual recognition based on deep learning in the visual grasping system of multi-joint manipulator can realize rapid multi-target recognition by training the deep convolution neural network without changing the recognition algorithm. In this paper, a depth neural network is built based on the idea of regression, and a reliable grasping position of a non-specific object is obtained from a given depth image, so that the multi-joint manipulator can further grasp. As an important part of machine learning, deep learning can establish an interactive relationship between multi-joint manipulator and complex dynamic environment.