👀 work in progress! many changes planned.
JAX-based robot library, focusing on modularity and ease of use.
We formulate goals and tasks as a nonlinear least squares problem, and use jaxls to solve it in a sparse manner.
Includes:
- Differentiable forward robot kinematics model, given a URDF from
yourdfpyas input.- Supports a wide range of robots, through robot-descriptions.
- Automatic robot collision geometry generation (e.g., with capsules).
- Differentiable collision bodies with numpy broadcasting logic, using a thin wrapper around MJX.
- Common cost factors (e.g., EE pose, self/world-collision, manipulability).
Supports:
- Arbitrary costs, as long as autodiff Jacobians are feasible.
pip install git+https://github.com/chungmin99/jaxmp.git
To run examples, install with pip install -e .[examples].