LeRobotLab

Personal exploration and experimentation with LeRobot datasets and models.

Project Overview

This repository contains experiments and analysis tools for working with LeRobot datasets from the HuggingFace Hub, with a focus on SO100 and SO101 robot datasets.

Quick Setup for Dataset Exploration

For just exploring datasets (this project's focus), you can use a minimal setup d

# Create conda environment
conda create -n lerobotlab python=3.10
conda activate lerobotlab

# Install dependencies
conda install jupyter pandas requests tqdm ipywidgets -c conda-forge
pip install huggingface_hub

The project folder should be next to a LeRobot install. For the LeRobot installation, follow the official installation guide: LeRobot Installation Instructions

Contents

SO100-SO101 Exploration Notebook

so100_so101_exploration.ipynb: Comprehensive search and analysis of SO100/SO101 robot datasets

• Searches HuggingFace Hub for relevant datasets
• Extracts metadata, licensing, and size information
• Generates clean CSV output for analysis
• Handles datasets with proper size estimation and error handling

Generated Output

• so100_so101_datasets_analysis.csv: Clean dataset analysis output with metadata and online links to view each dataset including videos.
• so100_so101_all_datasets.csv: Complete raw dataset information from HuggingFace Hub
• so100_so101_license_summary.csv: Summary of dataset licensing information

LeRobot Dataset Exploration Notebook

LeRobot_dataset_exploration.ipynb: Comprehensive exploration of all HuggingFace datasets authored by "lerobot"

• Discovers and processes metadata from all datasets under the lerobot/ namespace on HuggingFace Hub
• Covers a wide variety of robot hardware: ALOHA, Franka, UR5, Sawyer, Stretch, and many more
• Extracts detailed information: episodes, samples, FPS, robot types, camera configurations
• Downloads sample videos from datasets with camera data for local preview
• Generates interactive HTML visualization with embedded video players
• Creates organized folder structure for offline browsing
• Provides parallel processing for efficient dataset handling

Generated Output

• lerobot_datasets_videos.html: Interactive web interface for dataset exploration
• videos/: Local video samples organized by dataset and camera

About LeRobot

🤗 LeRobot is an open-source project providing models, datasets, and tools for real-world robotics in PyTorch. Learn more at huggingface.co/lerobot.

This project focuses specifically on exploring and analyzing the SO100/SO101 datasets available through the LeRobot ecosystem.