Grasping in cluttered scenes remains highly challenging for dexterous hands due to the scarcity of data. To address this problem, we present a large-scale synthetic benchmark, encompassing 1319 objects, 8270 scenes, and 427 million grasps. Beyond benchmarking, we also propose a novel two-stage grasping method that learns efficiently from data by using a diffusion model that conditions on local geometry. Our proposed generative method outperforms all baselines in simulation experiments. Furthermore, with the aid of test-time-depth restoration, our method demonstrates zero-shot sim-to-real transfer, attaining 90.7% real-world dexterous grasping success rate in cluttered scenes.
@article{arxiv.2410.23004,
title = {DexGraspNet 2.0: Learning Generative Dexterous Grasping in Large-scale Synthetic Cluttered Scenes},
author = {Jialiang Zhang and Haoran Liu and Danshi Li and Xinqiang Yu and Haoran Geng and Yufei Ding and Jiayi Chen and He Wang},
journal= {arXiv preprint arXiv:2410.23004},
year = {2024}
}