Skip to content

Sparsh

General tactile representations from self-supervised learning

Visit official site

What Sparsh is

The project is described as a set of general tactile representations trained by self-supervised learning for vision-based touch sensing.

What you can do with it

  • Train a representation with MAE, DINO or JEPA
  • Generate representations for several sensors
  • Work on force, slip and pose estimation

Tasks and resources

  • TacBench downstream tasks are named
  • Data-efficient training for new tasks is listed
  • PyTorch code and pretrained models ship together

Pros & cons

✓ What we like

  • Data-efficient training for new tasks is supported
  • Pretrained models and datasets are included

! What to watch out for

  • The material cites TacBench task results
  • Representations target vision-based tactile sensors

FAQ

Which sensors are named?

DIGIT, Gelsight'17 and Gelsight Mini are listed.

What comes with the project?

A PyTorch implementation, pretrained models and datasets are provided.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review