T-Rex: Tactile-Reactive Bimanual Dexterous Manipulation Dataset
- Provider
- University of California (T-Rex authors)
- License
- MIT (source)
- Size
- 7,903 files
- Format
- PNG, Parquet (in sampled files)
- Stored at
- Hugging Face (https://huggingface.co/datasets/zekaiwang/trex_dataset)
- Access
- Free and public. No account with the data provider is needed.
- Category
- Robotics
- Source
- https://arxiv.org/abs/2606.17055
About T-Rex: Tactile-Reactive Bimanual Dexterous Manipulation Dataset
T-Rex is a large tactile-reactive bimanual manipulation dataset collected by teleoperation on a Dexmate Vega-1 dual-arm robot fitted with two Sharpa Wave dexterous hands, each with five fingertip tactile sensors (arXiv:2606.17055). Manus gloves and VIVE trackers drive the robot; observations come from a head-mounted ZED X Mini camera and two wide-view ZED X One S wrist cameras, recorded at 640x360. Per fingertip the dataset stores the raw tactile sensor image and an estimated deformation map, both as video, and the estimated 6-axis net wrench. The card reports 5,464 episodes, 5,473,459 frames (about 50 hours at 30 fps), 5,370 language-annotated trajectories, 22 motor primitives and 207 objects. It is a LeRobot v3.0 dataset with Parquet under data/, MP4 under videos/ and an episodes_preview.parquet index. Measured 2026-11-14 at about 1.55 TB. Load with lerobot's LeRobotDataset, pandas or pyarrow; quickstart code is in the T-Rex GitHub repo. Useful for tactile-conditioned policies and contact-rich dexterous manipulation. Released under the MIT licence, copyright the Regents of the University of California; cite the T-Rex paper.
How to load T-Rex: Tactile-Reactive Bimanual Dexterous Manipulation Dataset in Python
from datasets import load_dataset
ds = load_dataset("zekaiwang/trex_dataset")
print(ds)Data structure
File types in sampled files: PNG, Parquet
Files
Example file: data/chunk-000/file-000.parquet
Analyze this dataset without downloading it
On Scigantic, T-Rex: Tactile-Reactive Bimanual Dexterous Manipulation Dataset opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.
This page is also available as Markdown and as JSON.
Related Robotics datasets
- Monado SLAM Datasets (MSD): Egocentric Visual-Inertial Recordings from Valve Index, HP Reverb G2 and Samsung Odyssey+ VR Headsets (Collabora)
- Video to Data: Robot Dexterity Task Library and Dataset
- PhysicalAI Robotics GR 00 T Teleop GR 1
- AgiBot World 2026: real-world dual-arm robot manipulation dataset
- PhysicalAI Autonomous Vehicle Cosmos Synthetic
- HiFi-UMI-2K: High-Fidelity Robot-Free Bimanual Manipulation
- DROID: In-the-Wild Franka Teleoperation (LeRobot port)
- AD Winter Driving Dataset: MI-Snow1000 Snow-Covered Lane Detection and MCity-WSPI Vehicle Dynamics on Snow and Ice (WMU AD Assurance Lab)