Skip to content

Repository files navigation

LeRobot Humanoid Workspace

This workspace groups the full LeRobot humanoid project in one place, from design to deployment:

  • mechanical/control co-design
  • hardware build assets
  • robot model assets
  • runtime and deployment stack
  • simulation-based identification

Each folder below is an independent Git repository with its own history/remotes.

Project Aim

Build an open and practical low-cost humanoid robot that people can reproduce.

Current design targets:

  • Cost target: less than USD 5,000 for the hardware (depending on sourcing, shipping, and taxes)
  • Lightweight structure: prioritize mass reduction and maintainability
  • 3D-printed first: most custom mechanical parts are printable and documented
  • Research-ready: reproducible stack from CAD to runtime and identification

Repository Names

Local folder names are aligned with canonical origin repository names:

  • lerobot-humanoid-design
  • lerobot-humanoid-hardware
  • lerobot-humanoid-identification
  • lerobot-humanoid-model
  • lerobot-humanoid-runtime

Repository Map

Folder Remote Role Main Outputs
lerobot-humanoid-design huggingface/lerobot-humanoid-design Mechanical/control co-design (URDF + optimization experiments) Design vectors, URDF assumptions, feasibility studies
lerobot-humanoid-hardware huggingface/lerobot-humanoid-hardware Build documentation, BOM, STL exports, wiring, commissioning tooling Printable parts, buy-list, assembly/electronics procedures
lerobot-humanoid-model huggingface/lerobot-humanoid-model Versioned model assets + Python helpers (lerobot_humanoid_models) MJCF/URDF assets and constants consumed by runtime/identification
lerobot-humanoid-runtime huggingface/lerobot-humanoid-runtime Real robot + simulation runtime, calibration, policy execution, LeRobot integration Deploy scripts, controllers, calibration flow, data acquisition
lerobot-humanoid-identification huggingface/lerobot-humanoid-identification MJWarp replay + CMA-ES parameter identification Identified simulator parameters and run artifacts

What Each Repo Is For

  • lerobot-humanoid-design: Define robot geometry and control assumptions early, then test feasibility with optimization and OCP experiments before hardware decisions.
  • lerobot-humanoid-hardware: Build the physical robot: BOM, STL files, print guidance, assembly steps, wiring, and motor commissioning.
  • lerobot-humanoid-model: Provide shared robot model assets (MJCF/URDF) and constants that other repos import.
  • lerobot-humanoid-runtime: Make the robot move in simulation and on real hardware (controllers, calibration, policy execution, deployment flow).
  • lerobot-humanoid-identification: Improve simulator realism by fitting parameters from real data with batched replay and CMA-ES.

How The Repos Connect

  1. design explores geometry/control tradeoffs and defines feasible robot assumptions.
  2. hardware is the build source of truth (what is physically assembled).
  3. model packages robot descriptions (MJCF/URDF) and constants for software stacks.
  4. runtime runs simulation and the real robot using calibrated models and policies.
  5. identification replays real logs to tune simulator parameters, feeding back into runtime/model/design.

Typical End-to-End Loop

  1. Update CAD/design assumptions (design, hardware).
  2. Export/update robot model assets (model).
  3. Validate calibration + policies in simulation/robot runtime (runtime).
  4. Collect logs and run joint-wise identification (identification).
  5. Push identified parameters back into runtime/model and iterate.

Quick Ops

Check state of every repo:

for repo in lerobot-humanoid-*; do
  echo "=== $repo ==="
  git -C "$repo" status -sb
  echo
 done

Show remotes for every repo:

for repo in lerobot-humanoid-*; do
  echo "=== $repo ==="
  git -C "$repo" remote -v | head -n 2
  echo
 done

Project Origin

Initially developed by Virgile Batto during his time at Hugging Face LeRobot.

About

No description, website, or topics provided.

Resources

Stars

238 stars

Watchers

5 watching

Forks

Releases

Packages

Used by

Contributors