Introduction

Codebase for replicating experiments in the NeurIPS 2020 paper "Memory-Efficient Learning of Stable Linear Dynamical Systems for Prediction and Control" by Giorgos Mamakoukas, Orest Xherija and Todd D. Murphey.

The master branch of this repository contains the code that we used to generate the results that appear on the paper. In the python branch, you will find a Python implementation of the SOC algorithm that we present in our paper, along with instructions on how to run it on your own data.

Table of contents

1. Datasets
2. Data Preparation
3. Dynamical Texture Experiments
4. Franka Emika Panda Experiments
5. Citing
6. Troubleshooting

Datasets

To get the datasets used for our experiments, read the instructions in the data directory.

Data Preparation

To prepare the datasets for the UCLA, UCSD and DynTex prediction experiments, follow the instructions in the prepare_data directory.

Dynamical Texture Experiments

To reproduce our results for the UCLA, UCSD and DynTex benchmarks, you will need to run the TrainDynamicTexture.m file.

NOTE: you will need to set some configuration options at the top of the TrainDynamicTexture.m file so that it can work on your particular system.

Franka Emika Panda Experiments

To reproduce our results from the simulations and experiments with the Franka Emika Panda robotic arm manipulator, consult the FrankaLDS directory.

Citing

If you find this project useful, consider

• starring this repository ⭐
• watching this repository for updates
• citing our paper (complete citation available after the publication of the NeurIPS 2020 proceedings)

@inproceedings{mamakoukas2020_memEfficientLDS,
  title={Memory-Efficient Learning of Stable Linear Dynamical Systems for Prediction and Control},
  author={Mamakoukas, Giorgos and Xherija, Orest and Murphey, Todd D.},
  booktitle={Advances in Neural Information Processing Systems 33},
  year={2020}
}

Troubleshooting

If you face any issues with our code or are unable to reproduce our results, please submit a Github issue and we will do our best to address it promptly.