HuggingFace

HuggingFace - LeRobot and the Open-Source Robotics Community

Overview

HuggingFace is the “GitHub” of AI models and datasets, bringing the open-source revolution to robotics through the LeRobot project. With SmolVLA and community datasets, they aim to make robotics AI accessible to everyone.

ItemDetails
HeadquartersNew York, NY / Paris, France
Founded2016
CEOClément Delangue
Robotics LeadRémi Cadène (Principal Research Scientist)
Core ProjectsLeRobot, SmolVLA

LeRobot

Project Introduction

LeRobot is an open-source project to democratize robotics AI.

🤗 LeRobot

"Making AI for Robotics more accessible
 with end-to-end learning"

Core Components

ComponentDescription
ModelsACT, Diffusion Policy, SmolVLA, etc.
DatasetsCommunity-shared datasets (continuously growing)
Hardware SupportSO100, Koch, Unitree G1, etc.
ToolsTraining, evaluation, visualization pipelines

Supported Hardware

  • SO100, SO101 (low-cost arms)
  • LeKiwi (mobile base)
  • Koch (bimanual)
  • Unitree G1 (humanoid)
  • HopeJR, OMX, EarthRover, and more

SmolVLA

Overview

A 450M parameter VLA released in June 2025:

ItemDetails
Parameters450M
FeatureRuns on MacBook
Training Data~30K episodes (community data)
ArchitectureVLM + Flow Matching

Performance

SmolVLA-450M ≈ Equivalent performance to 10× larger VLAs
             + Superior to ACT (LIBERO, Meta-World)
             + 30% faster with async inference

Efficiency

FactorFeature
TrainingTrainable on single GPU
InferenceRuns on consumer GPU/CPU
Fine-tuning~50 episodes, ~4 hours (A100)

Community Activities

LeRobot Hackathons

HackathonDetails
Oct 2024First community hackathon (Toulouse, France)
Jun 2025Worldwide hackathon (global simultaneous event)
OutcomesDiverse environment dataset sharing

Dataset Hub

HuggingFace Hub
└── lerobot/ (tag)
    ├── pusht_v2 (simulation)
    ├── aloha_static_* (ALOHA data)
    ├── so100_* (community collected)
    └── ... (many community datasets)

Community Contributions

  • Researchers and hobbyists worldwide sharing data
  • Diverse environments (labs, living rooms, warehouses)
  • Standardized formats for interoperability

Philosophy: Democratizing AI

Application to Robotics

LLM Democratization (HuggingFace Success)
├── Model Sharing → Hub
├── Training Tools → Transformers
└── Datasets → Datasets

Robotics Democratization (LeRobot Goal)
├── Model Sharing → lerobot/models
├── Training Tools → LeRobot library
└── Datasets → lerobot-tagged datasets

Emphasis on Low-Cost Hardware

EquipmentCost
SO100/SO101 arm~$100 (single)
LeKiwi base~$300
Teleoperation setup (2 arms)~$660

→ “Lowering barriers to entry for robotics research”


Roadmap

Below is compiled from public announcements and GitHub activity; may not be the official roadmap.

TimelineMilestone
2024LeRobot project launch
2024-25Community dataset growth
2025.06SmolVLA release
2025+More efficient models, broader hardware support (expected)

References


See Also