Control-Oriented Meta-Learning

This repository accompanies the papers:

• S. M. Richards, N. Azizan, J.-J. Slotine, and M. Pavone. "Control-oriented meta-learning". In International Journal of Robotics Research (IJRR), 2023. In press.
• S. M. Richards, N. Azizan, J.-J. Slotine, and M. Pavone. "Adaptive-control-oriented meta-learning for nonlinear systems". In Robotics: Science and Systems (RSS), 2021.

Getting started

Ensure you are using Python 3. Clone this repository and install the packages listed in requirements.txt. In particular, this code uses JAX.

Reproducing results

Training data can be generated with the commands ./generate pfar and ./generate pvtol.

Parameters can then be trained (for multiple training set sizes and random seeds) with the commands ./train pfar and ./train pvtol. This will take a while.

Test results can be reproduced with the commands ./test pfar and ./test pvtol. This may also take a while.

Finally, plots from the paper can be reproduced with commands ./plot pfar and ./plot pvtol.

Citing this work

Please use the following BibTeX entries to cite this work.

@ARTICLE{RichardsAzizanEtAl2023,
author    = {Richards, S. M. and Azizan, N. and Slotine, J.-J. and Pavone, M.},
title     = {Control-oriented meta-learning},
year      = {2023},
journal   = {International Journal of Robotics Research},
url       = {https://arxiv.org/abs/2204.06716},
doi       = {10.48550/arXiv.2204.06716},
note      = {In press},
}

@INPROCEEDINGS{RichardsAzizanEtAl2021,
author    = {Richards, S. M. and Azizan, N. and Slotine, J.-J. and Pavone, M.},
title     = {Adaptive-control-oriented meta-learning for nonlinear systems},
year      = {2021},
booktitle = {Robotics: Science and Systems},
url       = {https://arxiv.org/abs/2103.04490},
doi       = {10.15607/RSS.2021.XVII.056},
}