What it is
Large-scale vision-tactile-language-action robotics data with high-res tactile sensing for fine manipulation and dexterous control. 10,000+ planned hours of VTLA robot data; first 1,000 hours released across DM-DataClaw and DM-DataDex with RGB, tactile, pose/state, actions, audio, and language labels.
Key results
- 1,000 hours released; 10,000+ hours planned
- 2 platforms: DM-DataClaw and DM-DataDex
- 114-D state, 111-D action, CC-BY-NC-SA-4.0
Useful for training and evaluating tactile-aware policies for fine manipulation and dexterous skill learning.