A robotic grasp detection method based on auto-annotated dataset in disordered manufacturing scenarios

In the manufacturing industry, shorter product production cycles and steadily rising demand for customization put forward higher requirements for disordered grasping. The challenge of disordered grasping is the difficulty of labeling under disordered superposition and self-occlusion. Facing the challenge, the grasping angles classifi-cation model based on self-supervised learning is proposed, and an automatic annotation method is constructed for disordered grasp dataset. Based on the grasping angles classification model and the auto-annotated grasp dataset, a two-stage training, end-to-end using robotic grasp detection model is proposed, which can be used for robotic disordered grasping. The experiments and analyzes were carried out, and the results prove that the method is effective, and has strong generalization and high efficiency.