A Novel Real-time Grasping Method Cobimbed with YOLO and GDFCN

With the development of deep learning, the grasp detection of mobile robotic arms has been widely used in the fields of industrial automation and intelligent transportation. Most of the current grasp detection algorithms are aimed at relatively simple targets, and it is difficult to guarantee the grasp success rate of multi-modal targets in practical applications. In response to such problems, the algorithm in this paper combines and improves the object detection algorithm (YOLO v5) and the fully convolutional grasp detection algorithm (GDFCN), and proposes a real-time grasp detection algorithm for robotic arms suitable for unfamiliar objects in no training scenarios. In order to overcome the defect that single grasp detection cannot distinguish inherent objects and operable objects in the scene, this method first uses YOLO v5 to perform target detection, and then inputs the local depth image mapped by the target recognition frame into a lightweight fully convolutional neural network. In the network, feasible robot grasp detection is represented as diamond grasping, thus forming a new two-step cascade grasp algorithm. After experiments, the method in this paper can effectively complete the object classification task under the requirement of real-time performance, and enhance the stability and accuracy of grasp detection.