---
license: mit
---

# UniVTAC Benchmark Dataset

The UniVTAC Benchmark dataset provides simulation data for tactile-based robotic manipulation tasks.

## Overview

This dataset contains 100 episodes per task, totaling 800 episodes across 8 diverse manipulation tasks.

## Task Gallery

UniVTAC Benchmark currently includes the following manipulation tasks, all featuring tactile sensing:

| Task | Module | Description |
|---|---|---|
| **Collect** | `collect` | Collect contact-rich tactile data for pretraining |
| **Lift Bottle** | `lift_bottle` | Grasp and lift a bottle off a surface near a wall |
| **Lift Can** | `lift_can` | Grasp and lift a cylindrical can |
| **Insert HDMI** | `insert_HDMI` | Insert an HDMI connector into a port |
| **Insert Hole** | `insert_hole` | Precision peg-in-hole insertion |
| **Insert Tube** | `insert_tube` | Insert a tube into a fixture |
| **Pull Out Key** | `pull_out_key` | Extract a key from a lock |
| **Put Bottle in Shelf** | `put_bottle_in_shelf` | Place a bottle onto a shelf |
| **Grasp & Classify** | `grasp_classify` | Grasp an object and classify it by tactile feedback |

## Usage

For detailed instructions on data loading, environment setup, and benchmarking protocols, please visit our:  
👉 [official website](https://univtac.github.io/)
👉 [Github repo](https://github.com/univtac/UniVTAC)