Stanford Nimble

pip3 install nimblephysics-parco

Build a local wheel

If you want to use this checkout in another project, build and install the wheel:

python -m venv .venv-build
.venv-build/bin/python -m pip install --upgrade pip setuptools wheel build pybind11 pybind11-stubgen
.venv-build/bin/python setup.py bdist_wheel
.venv-build/bin/python -m pip install --force-reinstall dist/nimblephysics_parco-*.whl

If you are building from a fresh clone, populate .deps/ first. The build will recreate local helper symlinks such as Eigen and unsupported if your environment needs them, then you can rerun setup.py bdist_wheel.

Install from a wheel

For a local wheel file:

python -m pip install /path/to/nimblephysics_parco-*.whl

For a wheel published on GitHub Releases:

python -m pip install https://github.com/<org>/<repo>/releases/download/<tag>/nimblephysics_parco-*.whl

On Linux, make sure the runtime library paths used by the wheel are available on the target machine too.

If you're developing inside this repo, use an editable install:

python -m pip install -e .

** BETA SOFTWARE **

This repository provides a PyTorch-enabled physics engine with analytical backpropagation through simulation steps.

It also supports an analytical backwards pass, even through contact and friction.

It's as easy as:

from nimble import timestep

# Everything is a PyTorch Tensor, and this is differentiable!!
next_state = timestep(world, current_state, control_forces)