A Multimodal Grab Detection Network

In response to the problems of low accuracy and poor real-time performance of traditional robot grasping detection methods, a multi-modal robot grasping pose detection method is proposed. Design an end-to-end grasping detection network model and train it to output the grasping pose representation parameters of the target object. This network directly predicts suitable grasping positions by fusing RGB images and depth images.The experimental results show that on the Cornell publicly captured dataset, our model accuracy is 96.6%.