GraspNet-Atten:A Grasping Network Based on the Attention Mechanisms

This paper presents a grasping network based on an attention mechanism (GraspNet-Atten) for predicting robotic grasping. The model effectively captures key feature areas related to grasping direction in images by incorporating attention modules within the convolutional neural network architecture, thereby improving the accuracy and robustness of grasping predictions. Compared to the previous grasping detection network GGCNN, our method is better equipped to handle objects in complex scenarios, such as lighting effects and noisy environments, enhancing the model’s adaptability in diverse environments. Experimental results indicate that our proposed model achieves an accuracy of 83% on the publicly available Cornell Grasping Dataset, demonstrating superior performance compared to the GGCNN.