An Robot Vision Grasping Network Based on Inception-Lite

Abstract

Abstract Using computer vision algorithms to find the grasping position of unknown objects is a key issue in the development of intelligent robot technology. In order to address the problem of robot grasping, firstly, based on the analysis of Inception-v1, a novel Inception-Lite convolution module is proposed, and then a grasping network GRASPNET-CLF is designed based on the convolution module. Finally, the grasping network GRASPNET-CLF is trained and verified by GPU on Cornell grasping dataset. The experimental results show that the GRASPNET-CLF has better performance in the same kind of network, and meets the real-time requirements.