2019/03/06 by A. Gariépy, Gariépy, Alexandre, Jean-Christophe Ruel +5
Engineering · Medicine · #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Muscle activation and electromyography studies #Robot Manipulation and Learning #Robotics (cs.RO) #Stroke Rehabilitation and Recovery
paper · pdf · doi:10.48550/arxiv.1903.02489
openalex publication_date 2019/03/06 · openalex created_date 2022/07/29 · openalex updated_date 2026/07/28
Grasping is a fundamental robotic task needed for the deployment of household\nrobots or furthering warehouse automation. However, few approaches are able to\nperform grasp detection in real time (frame rate). To this effect, we present\nGrasp Quality Spatial Transformer Network (GQ-STN), a one-shot grasp detection\nnetwork. Being based on the Spatial Transformer Network (STN), it produces not\nonly a grasp configuration, but also directly outputs a depth image centered at\nthis configuration. By connecting our architecture to an externally-trained\ngrasp robustness evaluation network, we can train efficiently to satisfy a\nrobustness metric via the backpropagation of the gradient emanating from the\nevaluation network. This removes the difficulty of training detection networks\non sparsely annotated databases, a common issue in grasping. We further propose\nto use this robustness classifier to compare approaches, being more reliable\nthan the traditional rectangle metric. Our GQ-STN is able to detect robust\ngrasps on the depth images of the Dex-Net 2.0 dataset with 92.4 % accuracy in a\nsingle pass of the network. We finally demonstrate in a physical benchmark that\nour method can propose robust grasps more often than previous sampling-based\nmethods, while being more than 60 times faster.\n