jnsungp/unitree-g1-robocasa-pick-apple-bowl-contact-1k

---
license: cc-by-4.0
task_categories:
- robotics
tags:
- unitree-g1
- pick-and-place
- simulation
- curobo
- contact-force
- tactile
size_categories:
- 100K<n<1M
language:
- en
pretty_name: Unitree G1 Apple Pick and Place with Contact Force Dataset
---

# Unitree G1 Apple Pick and Place with Contact Force Dataset

<div align="center">
<table>
<tr>
<td align="center" style="vertical-align: top; padding: 5px;">
<img src="https://huggingface.co/datasets/jnsungp/unitree-g1-robocasa-pick-apple-bowl-contact-1k/resolve/main/frontview.png" width="500px" alt="Front View"/>
<br><b>Front View (Global)</b>
</td>
<td align="center" style="vertical-align: top; padding: 5px;">
<img src="https://huggingface.co/datasets/jnsungp/unitree-g1-robocasa-pick-apple-bowl-contact-1k/resolve/main/sideview.png" width="500px" alt="Side View"/>
<br><b>Side View (Profile)</b>
</td>
</tr>
<tr>
<td align="center" style="vertical-align: top; padding: 5px;">
<img src="https://huggingface.co/datasets/jnsungp/unitree-g1-robocasa-pick-apple-bowl-contact-1k/resolve/main/birdview.png" width="500px" alt="Top-Down View"/>
<br><b>Top-Down View (Bird's Eye)</b>
</td>
<td align="center" style="vertical-align: top; padding: 5px;">
<img src="https://huggingface.co/datasets/jnsungp/unitree-g1-robocasa-pick-apple-bowl-contact-1k/resolve/main/rs_view.png" width="500px" alt="Ego-Centric View"/>
<br><b>Ego-Centric View (Robot POV)</b>
</td>
</tr>
<tr>
<td colspan="2" align="center" style="padding-top: 20px;">
<i>Multi-view perspectives of the Unitree G1 performing the pick-and-place task.</i>
</td>
</tr>
</table>
</div>

## Dataset Description

The **Unitree G1 Apple Pick and Place with Contact Force Dataset** contains **968 high-quality trajectories** with **contact force measurements** from dexterous hands. The robot picks up a red apple and places it into a bowl using bilateral arms and tri-finger hands. All trajectories include 8-dimensional contact force data from finger and palm sensors.

**Key Features:**
- 968 successful trajectories with contact force data
- **8D contact force per timestep** (bilateral hand sensing)
- 28-DOF control: bilateral arms (7+7) + dexterous hands (7+7)
- 256×256 RGB video at 20 FPS (ego view)
- CuRobo motion planning (collision-free trajectories)
- MuJoCo + RoboCasa simulation with contact dynamics

**This dataset extends the base dataset with tactile/contact information for force-aware manipulation research.**

## Dataset Owner

**Junsung Park** ([@jnsungp](https://huggingface.co/jnsungp))
Seoul National University
[e-mail]

## License

Creative Commons Attribution 4.0 International (CC BY 4.0)

## Dataset Format

| Modality | Type | Shape | Description |
|----------|------|-------|-------------|
| **Observation State** | `float32` | `(28,)` | Joint positions (radians) for arms + hands |
| **Observation Contact** | `float32` | `(8,)` | **Contact forces (N) for bilateral hands** |
| **Action** | `float32` | `(28,)` | Target joint positions |
| **Video** | RGB | `(256, 256, 3)` | Ego view, 20 FPS, H.264 |
| **Language** | `string` | - | _"Pick up the red apple and place it on the bowl"_ |

### Contact Force Configuration (8D)

Contact force magnitudes measured at each finger and palm contact point:

| Index | Contact Point | Description |
|-------|---------------|-------------|
| 0 | `left_thumb` | Left hand thumb tip force |
| 1 | `left_index` | Left hand index fingertip force |
| 2 | `left_middle` | Left hand middle fingertip force |
| 3 | `left_palm` | Left hand palm contact force |
| 4 | `right_thumb` | Right hand thumb tip force |
| 5 | `right_index` | Right hand index fingertip force |
| 6 | `right_middle` | Right hand middle fingertip force |
| 7 | `right_palm` | Right hand palm contact force |

**Force Measurement:**
- **Units:** Approximate Newtons (N)
- **Method:** Computed from MuJoCo contact penetration depth
- **Formula:** `force ≈ max(0, -penetration_depth) × 100.0`
- **Range:** 0 (no contact) to ~50N (strong grasp)
- **Use Cases:** Grasp quality assessment, force-aware manipulation, slip detection

### Joint Configuration (28-DOF)

| Body Part | DOF | Description |
|-----------|-----|-------------|
| **Left Arm** | 7 | Shoulder (3) + Elbow (1) + Wrist (3) |
| **Right Arm** | 7 | Shoulder (3) + Elbow (1) + Wrist (3) |
| **Left Hand** | 7 | Index (2) + Middle (2) + Thumb (3) |
| **Right Hand** | 7 | Index (2) + Middle (2) + Thumb (3) |

## Dataset Statistics

- **Trajectories:** 968
- **Total Frames:** 278,772
- **Avg Episode Length:** ~288 frames (~14.4 seconds)
- **Episode Length Range:** 180-400 frames
- **Storage Size:** ~1.3 GB
- **Success Rate:** 100%

## Download

```bash
huggingface-cli download \
--repo-type dataset jnsungp/unitree-g1-robocasa-pick-apple-bowl-contact-1k \
--local-dir ./datasets/g1-contact
```

### Using Python

```python
from datasets import load_dataset

dataset = load_dataset("jnsungp/unitree-g1-robocasa-pick-apple-bowl-contact-1k")
```

## Dataset Structure

```
dataset_contact_force_1k/
├── data/
│ └── chunk-000/
│ ├── episode_000000.parquet
│ ├── episode_000001.parquet
│ └── ...
├── videos/
│ └── chunk-000/
│ └── observation.images.ego_view/
│ ├── episode_000000.mp4
│ ├── episode_000001.mp4
│ └── ...
├── meta/
│ ├── info.json # Dataset metadata
│ ├── stats.json # Statistics (mean, std, min, max)
│ ├── tasks.jsonl # Task descriptions
│ └── episodes.jsonl # Episode information
└── README.md
```

## Loading Data Example

```python
import pandas as pd
import numpy as np
import cv2

# Load trajectory with contact forces
df = pd.read_parquet("data/chunk-000/episode_000000.parquet")

# Access data
observations = df['observation.state'].values # (N, 28) - joint positions
contact_forces = df['observation.contact'].values # (N, 8) - contact forces
actions = df['action'].values # (N, 28) - target positions

# Analyze contact forces
left_thumb_force = contact_forces[:, 0] # Left thumb contact
right_palm_force = contact_forces[:, 7] # Right palm contact

# Check grasp strength at timestep t
t = 100
total_grasp_force = np.sum(contact_forces[t])
print(f"Total grasp force at t={t}: {total_grasp_force:.2f} N")

# Load video
cap = cv2.VideoCapture("videos/chunk-000/observation.images.ego_view/episode_000000.mp4")
```

## Use Cases

### 1. Force-Aware Manipulation
Train policies that use contact force feedback for:
- Adaptive grasping (adjust grip based on force)
- Fragile object handling
- Slip detection and prevention

### 2. Tactile Learning
- Learn force patterns for successful grasps
- Predict object properties from contact forces
- Multi-modal learning (vision + touch)

### 3. Sim-to-Real Transfer
- Fine-tune models with realistic force profiles
- Domain adaptation with contact dynamics
- Safety-aware manipulation

## Technical Details

**Simulation:**
- Platform: MuJoCo + RoboCasa
- Robot: Unitree G1 (upper body)
- Hands: Dex31 tri-finger hands with contact sensors
- Physics: Realistic contact dynamics and friction

**Motion Planning:**
- CuRobo (GPU-accelerated)
- Collision-free trajectories
- Smooth cubic interpolation

**Contact Sensing:**
- Simulated contact forces from MuJoCo
- Per-finger and palm measurements
- Force computed from penetration depth

## Comparison with Base Dataset

| Feature | Base Dataset | **Contact Force Dataset** |
|---------|--------------|---------------------------|
| Trajectories | 957 | **968** |
| Joint State | ✓ (28D) | ✓ (28D) |
| **Contact Force** | ✗ | **✓ (8D)** |
| Video | ✓ | ✓ |
| Use Case | Vision-based manipulation | **Force-aware manipulation** |

## Citation

```bibtex
@dataset{park2025unitree_g1_contact,
title={Unitree G1 Apple Pick and Place with Contact Force Dataset},
author={Park, Junsung},
year={2025},
publisher={Hugging Face},
url={https://huggingface.co/datasets/jnsungp/unitree-g1-robocasa-pick-apple-bowl-contact-1k}
}
```

## Acknowledgments

Built with [CuRobo](https://curobo.org/), [RoboCasa](https://robocasa.ai/), [MuJoCo](https://mujoco.org/), and Unitree G1.

---

**Version:** 1.0 | **Last Updated:** November 19, 2025