---
license: mit
task_categories:
- robotics
- imitation-learning
tags:
- lerobot
- imitation-learning
- vla
- open-x-embodiment
- action-chunking
- robotics
- manipulation
- teleoperation
- unscrewing
- mug-lid
- real-robot
size_categories:
- n<1K
---

# Unscrewing Travel Mug Lid Dataset

**Total Hours:** 5 seconds 
**Total Frames:** 288 
**Number of Episodes:** 1

![Preview](preview.gif)

## Dataset Description

This dataset contains synchronized `.mp4` videos and `.h5` data files for the task of unscrewing a travel mug lid. It is designed for training and evaluating robot learning models, particularly in imitation learning and reinforcement learning settings.

## Quickstart

```python
from datasets import load_dataset

ds = load_dataset("your-username/unscrewing-travel-mug-lid")
print(ds)
```

## Hardware & Collection Specs

- **Camera Model:** Iphone 16 Pro 
- **Resolution:** 4K 
- **FPS:** 60
- **Robot Arm Model:** None 
- **Provenance:** None

## Diversity & Data Splits

- **Number of Demonstrations:** 1 
- **Number of Objects/Variations:** 1
- **Number of Environments:** 1
- **Splits:**  
  - Train: [TRAIN_SPLIT]  
  - Validation: [VAL_SPLIT]  
  - Test: [TEST_SPLIT]

## Data Dictionary

The `.h5` files contain the following structure:

| Key | Type | Description |
| --- | --- | --- |
| `observations/images` | dict | Camera images (e.g., `cam0`, `cam1`) |
| `observations/state` | array | Robot joint states (angles, velocities) |
| `actions` | array | Executed actions (joint torques or positions) |
| `rewards` | array | Reward signal (if applicable) |
| `dones` | array | Episode termination flags |

## Intended Use Cases & Limitations

**Intended for:**
- Training imitation learning policies (e.g., Behavior Cloning, Action Chunking with Transformers)
- Fine-tuning Vision-Language-Action (VLA) models
- Evaluating robot manipulation algorithms

**Limitations:**
- The dataset may not cover all possible variations of unscrewing motions.
- The action space may be specific to the robot used.
- The data may contain biases from the collection environment.

## Action Space Definition

**Actions are defined as:** [ABSOLUTE or DELTA]  
**Coordinate Frame:** [COORDINATE_FRAME]  
**Units:** [UNITS]

## Failure Trajectories

**Contains failed attempts or human interventions:** NO  
**Details:** [DETAILS]

## Community & Support

For questions or support, please contact: hello@sonictech.com