facebookresearch/sparsh

Analysis updated 2026-07-17 · repo last pushed 2025-02-27

★ 228Jupyter NotebookAudience · researcherComplexity · 4/5StaleSetup · hard

Mindmap

mindmap
  root((repo))
    What it does
      Learns from touch data
      Trains feature extractor
      No labels needed
    Tech stack
      Jupyter Notebook
      Python
      MAE and DINO
    Use cases
      Detect gripper slip
      Estimate applied force
      Assess grasp stability
      Classify textiles
    Audience
      Robotics researchers
      Tactile sensing engineers
    Setup
      Download pretrained weights
      Fine tune on new task
      Live force demo

mindmap root((repo)) What it does Learns from touch data Trains feature extractor No labels needed Tech stack Jupyter Notebook Python MAE and DINO Use cases Detect gripper slip Estimate applied force Assess grasp stability Classify textiles Audience Robotics researchers Tactile sensing engineers Setup Download pretrained weights Fine tune on new task Live force demo

What do people build with it?

USE CASE 1

Fine-tune Sparsh's pretrained tactile representations to detect when a robot gripper is slipping.

USE CASE 2

Estimate the force applied at a contact surface using a DIGIT, GelSight 2017, or GelSight Mini sensor.

USE CASE 3

Train a slip-detection or grasp-stability model with far less labeled data than training from scratch.

USE CASE 4

Visualize real-time force fields from a DIGIT tactile sensor using the included live demo.

What is it built with?

Jupyter NotebookPythonPyTorch

How does it compare?

	facebookresearch/sparsh	facebookresearch/unibench	krishnaik06/text-summarization-nlp-project
Stars	228	228	198
Language	Jupyter Notebook	Jupyter Notebook	Jupyter Notebook
Last pushed	2025-02-27	2026-06-18	2024-08-17
Maintenance	Stale	Active	Stale
Setup difficulty	hard	moderate	hard
Complexity	4/5	3/5	4/5
Audience	researcher	researcher	developer

Figures from each repo's GitHub metadata at analysis time.

How do you get it running?

Difficulty · hard Time to first run · 1day+

Requires downloading large pretrained weights and datasets, plus a compatible tactile sensor for real-world use.

Not specified in the explanation.

Copy-paste prompts

Prompt 1

How do I download Sparsh's pretrained tactile representation weights and fine-tune them for force estimation on my DIGIT sensor data?

Prompt 2

Explain how Sparsh uses self-supervised methods like MAE and DINO to learn from unlabeled tactile sensor images.

Prompt 3

How do I reproduce Sparsh's slip-detection benchmark using the labeled downstream task datasets included in the repo?

Prompt 4

Walk me through running Sparsh's live demo that visualizes force fields from a DIGIT sensor in real time.

Frequently asked questions

What is sparsh?

A self-supervised learning framework that trains a reusable AI model to understand tactile sensor images, without needing labeled touch data.

What language is sparsh written in?

Mainly Jupyter Notebook. The stack also includes Jupyter Notebook, Python, PyTorch.

Is sparsh actively maintained?

Stale — no commits in 1-2 years (last push 2025-02-27).

What license does sparsh use?

Not specified in the explanation.

How hard is sparsh to set up?

Setup difficulty is rated hard, with roughly 1day+ to a first successful run.

Who is sparsh for?

Mainly researcher.

Open on GitHub → Explain another repo

This repo across BitVibe Labs

Verify against the repo before relying on details.