metadata
dataset_info:
dataset_name: SmolVLA Test Dataset
dataset_type: LeRobot
version: 2.1
SmolVLA Test Dataset
This is a test dataset created by the refactored SmolVLA library to demonstrate proper LeRobot format support.
Features
- LeRobot v2.1 format: Fully compatible with LeRobot training pipelines
- Episode structure preserved: Maintains temporal coherence
- Videos included: MP4 files for visual observations
- Proper metadata: Statistics, episode info, and task descriptions
Dataset Structure
.
├── data/ # Parquet files with trajectory data
├── videos/ # MP4 files for camera observations
└── meta/ # Metadata (info.json, episodes.jsonl, etc.)
Usage
from lerobot.datasets import LeRobotDataset
# Load the dataset
dataset = LeRobotDataset("SmolVLADataset/claude")
# Access samples
sample = dataset[0]
print(sample.keys())
Source
This dataset was created by merging LeRobot datasets using the SmolVLA library: https://github.com/smolvladataset/smolvladataset
The refactored library now properly:
- Downloads complete LeRobot datasets with videos
- Preserves episode structure and metadata
- Returns LeRobot-compatible dataset objects
- Validates dataset compatibility before merging