MJINX

Open in Colab PyPI version PyPI downloads

MJINX is a high-performance library for differentiable inverse kinematics, powered by JAX and MuJoCo MJX. The library was inspired by the Pinocchio-based tool PINK and Mujoco-based analogue MINK.

KUKA arm example GO2 robot example Heart path example Heart path example

Key Features

  1. Flexibility: Each control problem is assembled via Components, which enforce desired behavior or keep the system within a safety set.

  2. Multiple solution approaches: JAX’s efficient sampling and autodifferentiation enable various solvers optimized for different scenarios.

  3. Fully JAX-compatible: Both the optimization problem and solver support JAX transformations, including JIT compilation and automatic vectorization.

  4. Convenience: The API is designed to make complex inverse kinematics problems easy to express and solve.

Citing MJINX

If you use MJINX in your research, please cite it as follows:

@software{mjinx25,
author = {Domrachev, Ivan and Nedelchev, Simeon},
license = {MIT},
month = mar,
title = {{MJINX: Differentiable GPU-accelerated inverse kinematics in JAX}},
url = {https://github.com/based-robotics/mjinx},
version = {0.1.1},
year = {2025}
}

Contents: