OmniDex: Scaling Dexterous Hand Grasping to Diverse Cluttered Scenes
OmniDex: Scaling Dexterous Hand Grasping to Diverse Cluttered Scenes
This paper presents OmniDex, a model that addresses the challenge of dexterous grasping in cluttered environments by curating a large-scale benchmark with over 2.6 million scenes and 0.4B grasp ground truths. The model combines Soft Winner-Takes-All learning with human-inspired physical constraints to achieve robust grasping without post-optimization latency, demonstrating strong generalization across diverse scenes and objects.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
scenes
>2,600,000
grasp ground truths
400,000,000
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
is, we curate high-quality 3D objects and supporting bases, proposing a scalable seed-and-filter strategy that bypasses sluggish scene-level optimization. This yields an unprecedented benchmark comprising over 2.6 million scenes and 0.4B scene-specific grasp ground truths, featuring diverse realistic layouts paired with rich semantic and geometric observations. Furthermore, we introduce the OmniDex model to overcome
Open source S3
Source：arXiv Robotics — research abstracts · arxiv.org