Multimodal Deep Learning for Robotic Grasping

Application Background: This code is designed for robotic grasp learning, enabling robots to determine optimal grasping strategies (e.g., grasping a cup by its body vs. handle). It utilizes multimodal data including cups, remote controls, and cameras to train neural networks in grasp decision-making. Implemented in MATLAB with detailed post-extraction instructions. Key Technologies: Includes a comprehensive README file detailing usage procedures and downloadable training datasets. Core training is executed through trainGraspRecMultiSparse.m in recTraining folder, implementing sparse multimodal network architecture for grasp preference learning.