prpl-kinematics

Kinematics-only robot modeling, IK, motion planning, and manipulation primitives built on a general KinematicTree. A ground-up successor to pybullet-helpers: engine-agnostic (PyBullet is a pluggable collision/render backend, not the source of truth), with poses represented as spatialmath SE3/SE2.

See DESIGN.md for the architecture and roadmap.

Status

Early, and the API may still change. The geometry layer and the KinematicTree (joints, forward kinematics, grasp via re-parenting, state snapshots) are the stable core. Built on top of them: URDF loading, PyBullet and Blender backends, numerical and analytic IK (IKFast, EAIK, optionally SSIK), BiRRT and OMPL motion planning, manipulation primitives, and robot models for the Franka Panda, Kinova Gen3, TidyBot, and the bimanual Dexmate Vega.

Pin a version if you need stable behavior.

Requirements

• Python 3.10+

Installation

pip install prpl_kinematics

IKFast solvers are compiled from the bundled C++ sources on first use, which needs a C++ toolchain and LAPACK/BLAS. The other IK backends have no such requirement.

Optional extras

pip install "prpl_kinematics[planning]"   # OMPLPlanner and seed_ompl
pip install "prpl_kinematics[ssik]"       # the SSIK analytic IK backend

ompl publishes wheels for far fewer platforms than everything else here (none for Windows, and macOS 15 or newer only), so it stays out of the base install. Importing OMPLPlanner or seed_ompl without it raises an ImportError saying so; everything else, including BiRRTPlanner, works untouched.

Development

From the monorepo root, with uv:

uv pip install -e "./prpl-kinematics[develop]"

./run_ci_checks.sh   # autoformat, mypy, pylint, pytest