Touch and Go: Paired Vision-Tactile Dataset
Paired visual and tactile data for cross-modal robot perception. Learn to predict touch from sight and sight from touch.
Overview
Touch and Go provides synchronized visual images and tactile sensor readings from interactions with diverse objects and surfaces. The dataset is designed for training cross-modal representations -- models that can predict what something feels like from its appearance, and vice versa. This capability is essential for robots that need to plan grasps and manipulation strategies based on visual input but execute them with tactile feedback.
The dataset covers a variety of materials, textures, and object shapes, providing a foundation for material recognition, hardness estimation, and slip detection from visual cues alone.
Access
Note: This dataset is released for research use only. Commercial use is not permitted.
Related datasets
- GelSight Tactile -- vision-based tactile sensing data
- RH20T -- multi-modal dataset including tactile at 200Hz
- DROID -- large-scale manipulation (visual only)