PyRoki

UC Berkeley's open-source JAX toolkit for solving robot kinematics optimization problems.

PyRoki is a Python toolkit open-sourced in 2025 by a team at UC Berkeley, built on JAX (Google's numerical computing library that supports automatic differentiation and compiling to GPU). It formulates problems like inverse kinematics, trajectory optimization, and motion retargeting uniformly as nonlinear optimization over “cost terms plus constraints”: users combine modules for collision avoidance, joint limits, end-effector pose, and so on as needed, and the solver runs on CPU or GPU with support for batched computation. The authors report that it's faster than some existing tools for inverse kinematics. It's often used alongside Viser, a visualization tool from the same group, and is well suited to research prototypes such as retargeting human hand or body motion onto a robot, or generating training data.

ExampleTake wrist trajectories estimated from a video of a human demonstration, and use PyRoki to optimize them into joint trajectories for a bimanual robot while avoiding self-collision.

Also called

Python Robot Kinematics

Related

Inverse Kinematics (IK) · Trajectory Optimization · Motion Retargeting · JAX · Viser · Pink

Sources

chungmin99/pyroki (GitHub)
PyRoki: A Modular Toolkit for Robot Kinematic Optimization (arXiv)

As of

2025-05