metadata
license: mit
task_categories:
- robotics
language:
- en
tags:
- robotics
- isaac-lab
- cross-embodiment
- manipulation
- franka
- stretch
- fetch
size_categories:
- 10K<n<100K
🤖 Robot Zoo
An open-source multi-embodiment robotics dataset for cross-embodiment research. Built with Isaac Lab 0.54.3 + Isaac Sim 5.1.0.
Demo Video
▶️ Watch on YouTube | Download MP4
Dataset Summary
| Task | Franka | Stretch | Fetch | Total |
|---|---|---|---|---|
| Pick Cube | 1000 | 1058 | 1000 | 3058 |
| Open Drawer | 1000 | - | - | 1000 |
| Reach | 1000 | 1000 | 1000 | 3000 |
| Place Cube | 1000 | 1000 | 1000 | 3000 |
| Push | 1000 | - | 1000 | 2000 |
| Total | 5000 | 3058 | 4000 | 13,058 |
Robots
| Robot | Type | DOF | Base |
|---|---|---|---|
| Franka Panda | Fixed arm | 7 | Fixed |
| Stretch 3 | Mobile manipulator | 8 | Mobile |
| Fetch | Mobile manipulator | 7 | Mobile |
Data Format
Each episode is saved as a .npz file with keys:
- observation_state — joint positions and velocities
- observation_cube_pos — object/target position (3D)
- action — commanded joint targets
- reward — per-step reward signal
How to Load
from huggingface_hub import snapshot_download
import numpy as np
from pathlib import Path
path = snapshot_download(repo_id="shareef14/robot-zoo", repo_type="dataset")
ep = np.load(Path(path) / "fetch_pick_cube" / "episode_0000.npz")
print(ep["observation_state"].shape)
print(ep["observation_cube_pos"].shape)
print(ep["action"].shape)
GitHub
https://github.com/Shareefbaba/robot-zoo
Author
Shareef Baba Shaik — Robotics Engineer, Hyderabad, India
GitHub · LinkedIn · Portfolio · HuggingFace