Constrained Manifold Neural Motion Planning with B-splines (CNP-B)

Implementation of P. Kicki et al., "Fast Kinodynamic Planning on the Constraint Manifold With Deep Neural Networks," in IEEE Transactions on Robotics, vol. 40, pp. 277-297, 2024.

See also:
paper
website
preprint

Dependencies

General

• Tensorflow pip install tensorflow
• Tensorflow-graphics pip install tensorflow-graphics
• NumPy pip install numpy
• Matplotlib pip install matplotlib
• Pinocchio sudo apt install ros-noetic-pinocchio

For Air-Hockey hitting

• Nlopt sudo apt install libnlopt-cxx-dev libnlopt-dev
• Coin-or-CLP sudo apt install coinor-libclp-dev

For demonstration of motion planning in ROS

• docker

For results plotting and statistical analysis

• SciPy ``
• statsmodels pip install statsmodels

Usage

Download data

bash download_datasets.sh

Download pre-trained models

bash download_models.sh

Build python bindings (for Air Hockey hitting only)

bash build.sh

Make an inference of the model on a sample Air Hockey hitting problem

python examples/air_hockey_hitting.py

or moving a vertically oriented heavy object

python examples/heavy_object.py

Use for motion planning in ROS

Run docker container

cd docker && ./run.sh

Run demo

bash demo/demo.sh

Cite

@ARTICLE{kicki2024kinodynamic,
  author={Kicki, Piotr and Liu, Puze and Tateo, Davide and Bou-Ammar, Haitham and Walas, Krzysztof and Skrzypczyński, Piotr and Peters, Jan},
  journal={IEEE Transactions on Robotics}, 
  title={Fast Kinodynamic Planning on the Constraint Manifold With Deep Neural Networks}, 
  year={2024},
  volume={40},
  number={},
  pages={277-297},
  doi={10.1109/TRO.2023.3326922}}