# Dexora Real-World Dataset: High-DoF Bimanual Dexterous Teleoperation

## Facts

- Provider: Dexora team
- License: MIT (https://huggingface.co/datasets/Dexora/Dexora_Real-World_Dataset)
- Size: 100,000 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/Dexora/Dexora_Real-World_Dataset
- Access: public, readable without an account
- Category: Robotics
- Source: https://github.com/ZZongzheng0918/Dexora
- Page: https://scigantic.com/archive/3c576e3e-8437-5ec2-bf7e-61f5abd0e445
- Keywords: Robotics, robot learning, Parquet, Hugging Face

## About

The Dexora real-world dataset is a corpus of teleoperated bimanual dexterous manipulation released with Dexora, an open vision-language-action model for high-DoF bimanual dexterity. The card reports 12.2K episodes, 2.92 million frames and 40.5 hours, collected with a hybrid teleoperation system that uses an exoskeleton for the arms and an Apple Vision Pro for the hands to drive a 36-DoF bimanual robot. Each episode has synchronised RGB from an egocentric head camera, left and right wrist cameras and a static third-person camera, plus robot proprioception, actions and language instructions; the card says episodes follow the LIBERO-2.1 standard. Tasks span pick-and-place (about 55 percent), dexterous manipulation (20 percent, such as twisting caps and cutting leeks), assembly (15 percent) and articulated objects (10 percent), with 347 objects in 17 categories and a published object inventory. The repo has a task-level view under dexora/ (about 200 task folders) and category folders airbot_pick_and_place, airbot_assemble, airbot_articulation and airbot_dexterous. Measured 2026-11-14 at about 214 GB. The card carries an MIT licence badge; cite the Dexora project.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("Dexora/Dexora_Real-World_Dataset")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `airbot_articulation/data/chunk-000/episode_000000.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/3c576e3e-8437-5ec2-bf7e-61f5abd0e445
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/3c576e3e-8437-5ec2-bf7e-61f5abd0e445
- Over MCP: `get_schema_card` with id `3c576e3e-8437-5ec2-bf7e-61f5abd0e445` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Robotics datasets](https://scigantic.com/topics/robotics)
- [Robot learning datasets](https://scigantic.com/topics/robot-learning)
