Sparsh
facebookresearch
Self-supervised encoders for vision-based touch, with downstream evaluation through TacBench.
What you can explore
Explore tactile representations and the documented DIGIT and GelSight Mini examples.
Before you use it
Touch-image encoders do not directly provide calibrated force or slip guarantees. Check sensor preparation and each checkpoint’s intended input.
Original contributors
Carolina Higuera, Akash Sharma and collaborators
License / access: CC BY-NC 4.0 · check dataset and dependency terms
Editorial reference · checked 2026-09-22