C++ MIT

momentum

A library for human kinematic motion and numerical optimization solvers to apply human motion

F

facebookresearch

Dernière activité 28 sept. 2026
facebookresearch/momentum

417

étoiles

66

forks

7

issues ouvertes

Ce README est souvent en anglais.

Momentum

CI Windows CI macOS CI Ubuntu Publish Website PyPI Wheels

Momentum provides foundational algorithms for human kinematic motion and numerical optimization solvers to apply human motion in various applications.

Forward and Inverse Kinematics with Interpretable Parameterization RGBD Body Tracking Solver Monocular RGB Body Tracking Solver

Quick Start

Installation

Pre-built binaries are available for Windows, macOS, and Linux:

Install full conda/pixi packages -- recommended

# Install full python package from pixi, including modules with cpp torch dependencies
pixi add pymomentum

# Install full python package from conda, including modules with cpp torch dependencies
conda install -c conda-forge pymomentum

# Install cpp momentum package from pixi, including modules with cpp torch dependencies
pixi add momentum-cpp

# Install cpp momentum package from conda, including modules with cpp torch dependencies
conda install -c conda-forge momentum-cpp

Install PyPI wheels -- these are experimental ⚠️

# Install the NumPy/native core without PyTorch
pip install pymomentum-core

# Install core package with optional visualization helpers
pip install "pymomentum-core[viser,rerun]"

# Add diff_geometry, Torch helpers, and differentiable solvers for CPU PyTorch
pip install pymomentum-cpu

# Install full package including diff_geometry and differentiable solver modules, linked against CUDA PyTorch
pip install "torch>=2.8,<2.9" --index-url https://download.pytorch.org/whl/cu129
pip install pymomentum-gpu

The same layered packages are available from conda-forge. pymomentum remains a compatibility package that installs the CPU or GPU stack selected by the solver.

The pymomentum-core package owns the NumPy/native modules geometry, solver2, marker_tracking, axel, camera, and renderer. It does not install PyTorch. pymomentum-cpu and pymomentum-gpu are mutually exclusive add-ons: each installs the Torch helpers plus diff_geometry and solver, and depends on a compatible pymomentum-core release.

⚠️ PyPI support is experimental. For the most stable experience, we recommend using Conda or Pixi.

📦 Browse packages: conda-forge • prefix.dev • PyPI

Building from Source

git clone https://github.com/facebookresearch/momentum
cd momentum
pixi run build      # Builds C++ library and Python bindings
pixi run test       # Runs tests
pixi run hello_world  # Runs example

For detailed instructions, see the comprehensive guides on our website:

📖 Documentation

Visit our documentation website for comprehensive guides, examples, and API references:

Contributing

Check our contributing guide to learn about how to contribute to the project.

License

Momentum is licensed under the MIT License. A copy of the license can be found here.

Projets similaires

The Open Motion Planning Library (OMPL)

C++motion-planningrobotics
Oompl
2,2 k étoiles713

Motion, a software motion detector. Home page: https://motion-project.github.io/

C++
MMotion-Project
4 k étoiles560

Multi-Joint dynamics with Contact. A general purpose physics simulator.

C++mujocophysicsrobotics
Ggoogle-deepmind
15,4 k étoiles1,8 k