Holohover - SysID

This repository provides the code accompanying the paper On identifying the non-linear dynamics of a hovercraft using an end-to-end deep learning approach.

Getting Started

All configuration is done through the params.json file. It specifies the data source, learning parameters, and initial model parameters.

Preprocessing Data

Before learning a new mode, the data has to be preprocessed. To do so, create a new folder in the experiments folder, add the recorded mcap bag, change the data/experiment field in params.json, and run

python3 run_preprocessing.py

Learning a Model

Change the params.json file accordingly and run

python3 run_learning.py

Datasets and Models from the Paper

The data used for training is 2023_10_18-11_28_12_sysid_h1_old and the data used for validation is 2023_11_22-11_57_44_sysid_h1_old. The relevant models are located in models/paper and the RMSE calculation is done in paper_results.ipynb. The control experiment and analysis can be found in control_experiment.

Citing our Work

To cite our work in other academic papers, please use the following BibTex entry:

@misc{schwan2024,
author={Schwan, Roland and Schmid, Nicolaj and Chassaing, Etienne and Samaha, Karim and Jones, Colin N.},
title={On identifying the non-linear dynamics of a hovercraft using an end-to-end deep learning approach}, 
year={2024},
eprint = {arXiv:2405.09405},
}