Wheel-Chair-Planning

A planning library for the wheelchair-mounted xArm7 robot (wheelchair_xarm_description). It provides inverse kinematics (TRAC-IK and Pink), motion planning (VAMP + OMPL), and collision-aware planning through a unified Python interface.

The architecture mirrors AdaCompNUS/Autolife-Planning; only the robot has been swapped — from the 24-DOF Autolife humanoid to a 10-DOF mobile manipulator (a 3-DOF planar wheelchair base + a 7-DOF xArm7 arm).

Features

• Inverse Kinematics — TRAC-IK (unconstrained) and Pink (QP-based constrained) solvers with self-collision avoidance and camera/orientation stabilization
• Motion Planning — VAMP-based planner with collision checking, path validation, and subgroup planning (arm-only, base-only, and whole-body)
• Time Parameterization — TOPP-RA; converts planned paths into executable trajectories with velocity/acceleration limits
• Kinodynamic Planning — FLASK (flatness-based kinodynamic RRT-Connect) plans time-parameterised trajectories directly, with joint/base velocity and acceleration limits and an exactly nonholonomic diff-drive base
• Collision Geometry — Spherized URDF representations for efficient collision detection, with pointcloud obstacle support

Robot

Quick Start

Platform: Linux x86_64 (AVX2), Python 3.12–3.15.

For inference — running the planners and IK solvers — pip install:

git clone --recursive https://github.com/H-tr/Wheel-Chair-Planning.git
cd Wheel-Chair-Planning
pip install -e ".[kinematics]"   # core deps: numpy, scipy, casadi

The kinematics extra adds pin + pin-pink for the IK solvers; drop it if you only need motion planning and time parameterization.

For development — rebuilding URDFs, regenerating FK headers, running the C++ toolchain end-to-end — use the setup script, which installs pixi and the conda-forge deps (pinocchio, orocos-kdl, eigen, boost, ...):

bash scripts/setup.sh

Usage

# Inverse kinematics
pixi run python examples/ik/basic.py
pixi run -e dev python examples/ik/basic_vis.py           # PyBullet visualization

# Motion planning
pixi run python examples/planning/motion.py
pixi run python examples/planning/subgroup.py

# Time parameterization
pixi run python examples/planning/time_parameterization.py

# Kinodynamic planning (FLASK): drive to a table, then reach over it
pixi run python examples/planning/kinodynamic.py

# Tests
pixi run -e dev test

Rebuilding the robot description

The planning URDFs, spherized collision geometry, and the FK header are generated from the vendored robot description:

pixi run build-robot        # preprocess URDF + meshes + SRDF + virtual base
pixi run decompose-robot    # CoACD convex decomposition
pixi run spherize-robot     # FOAM spherization -> wheelchair_spherized.urdf
pixi run generate-fk        # cricket -> ext/ompl_vamp/robot/wheelchair.hh
pixi run build-pkg          # compile the C++ extensions

Project Structure

wheelchair_planning/   # Core Python package
  kinematics/          # TRAC-IK + Pink IK, FK, collision checking
  planning/            # VAMP motion planning, cost + constrained planners
  trajectory/          # TOPP-RA time parameterization
  envs/                # Simulation environments (PyBullet)
  types/               # Shared dataclasses (SE3Pose, PlannerConfig, ...)
  resources/           # Packaged URDFs and asset loaders
third_party/
  vamp/                # SIMD-accelerated motion planning (submodule)
  ompl/                # Open Motion Planning Library (submodule)
  cricket/             # FK code generator (submodule)
  foam/                # Collision geometry / spherization (submodule)
  toppra/              # Time-optimal path parameterization (submodule)
ext/                   # C++ extensions (ompl_vamp, trac_ik)
assets/
  wheelchair_xarm_description/   # Raw robot description (submodule)
examples/              # IK, planning, demos
tests/                 # Pytest suite (CI)
scripts/               # Setup, build, spherize, FK codegen
docs/                  # MkDocs site sources (GitHub Pages)

Acknowledgements

This project builds on several outstanding open-source libraries and on the AdaCompNUS/Autolife-Planning architecture:

• VAMP — SIMD-accelerated motion planning and collision checking (KavrakiLab, Rice University).
• OMPL — The Open Motion Planning Library (KavrakiLab, Rice University).
• TOPP-RA — Reachability-analysis-based time-optimal path parameterization, used for time parameterization.
• TRAC-IK — Inverse kinematics solver (TRACLabs).
• wheelchair_xarm_description — The robot description (URDF + meshes).