Port-Hamiltonian Neural ODE Networks on Lie Groups For Robot Dynamics Learning and Control

This repo provides code for our paper "Port-Hamiltonian Neural ODE Networks on Lie Groups For Robot Dynamics Learning and Control". Please check out our project website for more details: https://thaipduong.github.io/LieGroupHamDL/.

Dependencies

Our code is tested with Ubuntu 18.04 and Python 3.7, Python 3.8. It depends on the following Python packages:

torchdiffeq 0.1.1, torchdiffeq 0.2.3

gym 0.18.0, gym 1.21.0

gym-pybullet-drones: https://github.com/utiasDSL/gym-pybullet-drones

torch 1.4.0, torch 1.9.0

numpy 1.20.1

scipy 1.5.3

matplotlib 3.3.4

pyglet 1.5.27 (pendulum rendering not working with pyglet >= 2.0.0)

Notes: NaN errors might happen during training with torch 1.10.0 or newer due to numerical issues!!!!!!!!! Please switch to float64 to fix this.

Demo with pendulum

Run python ./examples/pendulum/train_pend_SO3_friction.py to train the model with data collected from the pendulum environment. It might take some time to train. A pretrained model is stored in ./examples/pendulum/data/run1_pend_friction/pendulum-so3ham_ode-rk4-5p.tar

Run python ./examples/pendulum/analyze_pend_SO3.py to plot the generalized mass inverse M^-1(q), the potential energy V(q), and the control coefficient g(q)

Run python ./examples/pendulum/comparison.py to verify that compared to other baselines, our framework learns better, respects energy conservation, SE(3) constraints, and the phase portrait of a trajectory rolled out from the learned dynamics.

Run python ./examples/pendulum/control_pend_SO3.py to test the energy-based controller with the learned dynamics.

Demo with quadrotor

Run python ./examples/quadrotor/train_quadrotor_SE3.py to train the model with data collected from the pybullet drone environment. It might take some time to train. A pretrained model is stored in ./examples/quadrotor/data/quadrotor-se3ham-rk4-5p.tar

Data collection

Run python ./examples/quadrotor/analyze_quadrotor_SE3.py to plot the generalized mass inverse M^-1(q), the potential energy V(q), and the control coefficient g(q)

Run python ./examples/quadrotor/comparison.py and python ./examples/quadrotor/comparison_plottraj.py to verify that compared to other baselines, our framework learns better, respects energy conservation and SE(3) constraints by construction.

Run python ./examples/quadrotor/control_quadrotor_SE3.py to test the energy-based controller with the learned dynamics.

Demo with PX4 simulated quadrotors

Please check out our branch "PX4" and follow instructions in README for demos with PX4 simulated quadrotors.

Data collection

https://github.com/thaipduong/LieGroupHamDL/assets/40247151/733f017d-4cc2-4f29-b020-fa753d3fffce

Trajectory tracking with learned dynamics

https://github.com/thaipduong/LieGroupHamDL/assets/40247151/bfd06715-bba2-42dc-b423-83c0cd5017ca

Citation

If you find our papers/code useful for your research, please cite our work as follows.

1. T. Duong, A. Altawaitan, J. Stanley, N. Atanasov. Port-Hamiltonian Neural ODE Networks on Lie Groups For Robot Dynamics Learning and Control. under review, 2023.

   @article{duong23porthamiltonian,
   author = {Thai Duong, Abdullah Altawaitan, Jason Stanley AND Nikolay Atanasov},
   title = {Port-{H}amiltonian Neural {ODE} Networks on Lie Groups For Robot Dynamics Learning and Control},
   journal = {arXiv preprint arXiv:2401.09520},
   year = {2023},
   }

2. T. Duong, N. Atanasov. Hamiltonian-based Neural ODE Networks on the SE(3) Manifold For Dynamics Learning and Control. RSS, 2021

   @inproceedings{duong21hamiltonian,
   author = {Thai Duong AND Nikolay Atanasov},
   title = {{Hamiltonian-based Neural ODE Networks on the SE(3) Manifold For Dynamics Learning and Control}},
   booktitle = {Proceedings of Robotics: Science and Systems},
   year = {2021},
   address = {Virtual},
   month = {July},
   DOI = {10.15607/RSS.2021.XVII.086} 
   }