DDGC: Generative Deep Dexterous Grasping in Clutter

Recent advances in multi-fingered robotic grasping have enabled fast 6-Degrees-of-Freedom (DOF) single object grasping. Multi-finger cluttered scenes, on the other hand, remains mostly unexplored due to added difficulty of reasoning over obstacles which greatly increases computational time generate high-quality collision-free grasps. In this work, we address such limitations by introducing DDGC, a generative multi-finger grasp sampling method that can high quality grasps scenes from RGB-D image. DDGC is built as network encodes scene information produce coarse-to-fine poses and configurations. We experimentally benchmark against two state-of-the-art methods 1200 simulated 7 real-world scenes. The results show outperforms baselines synthesizing removing clutter. also 4-5 times faster than GraspIt!. This, turn, opens door for using practical applications has so far been limited excessive computation needed methods. Code videos are available at https://irobotics.aalto.fi/ddgc/.