Motion planning is a computational problem to find a sequence of valid configurations that moves the object from the source to destination.

This library is designed for robotic applications, especially sampling-based algorithms for high-dimension configuration spaces(e.g., robot arm).

• pythonic: easy to debug, customize and extend
• standalone collision checker (hpp-fcl): without relying on any physical simulator (e.g., mujoco, pybullet, sapien) to check collision
• out-of-box: common motion planning algorithms (e.g., RRT-Connect) are implemented for robotic manipulation

This library (pymp) depends on pinocchio to handle URDF and robot kinematics, hpp-fcl to check collision, toppra to do time parameterization.

From pip:

pip install motion-planning

From source:

git clone https://github.com/Jiayuan-Gu/pymp.git
pip install -e .

From Github directly:

pip install --upgrade git+https://github.com/Jiayuan-Gu/pymp.git

See example.py for basic usage. Note that pymp depends on SRDF associated with URDF to remove self-collision.

The logging level can be specified by the environment variable PYMP_LOG.

# Set the logging level to DEBUG for pymp
export PYMP_LOG=DEBUG

pymp supports specifying the pose of the base link during the initialization of the planner. We support many formats of pose (e.g., [x, y, z] for position, [w, i, j, k] for quaternion, [x, y, z, w, i, j, k] for SE(3), or a 4x4 rigid transformation matrix)

from pymp import Planner

planner = Planner(
    ...
    base_pose=[0, 0, 0],
)