---
license: mit
task_categories:
- reinforcement-learning
- robotics
---

# Dataset Card for Sagar18/ball-maze-images

This dataset contains visual observations of a ball maze environment, derived from the original state-space dataset at `notmahi/tutorial-ball-top-20`.

## Dataset Description

- **Features**:
  - `image`: JPEG-encoded RGB images showing the ball maze environment
  - `state`: Original state vector [x, y, dx, dy] representing ball position and velocity
  - `action`: Actions taken in the environment
  - `episode_index`: Index of the episode
  - `frame_index`: Frame number within the episode
  - `next_done`: Whether the frame is terminal
  - `index`: Global index in the dataset

## Usage
```python
from datasets import load_dataset

# Load the dataset
dataset = load_dataset("Sagar18/ball-maze-images")

# Access images (they will be automatically decoded)
image = dataset[0]["image"]
state = dataset[0]["state"]
```