Research on Robotic Arm Grasp Detection Method Based on Multimodal Fusion

In order to improve the grasping success rate, this paper proposes a robot grasping detection neural network based on multimodal fusion. The main contents are as follows: First, the grasping detection task based on neural network is introduced. Then, a MESB-GDN is proposed. Then, by conducting grasping detection experiments on Cornell dataset and Jacquard dataset, it is proved that the proposed MESB-GDN has better prediction performance than other SOTA methods. Finally, a physical grasping experiment is carried out on the robot, which proves that the robot using MESB-GDN has the highest autonomous grasping success rate compared with other SOTA methods.