Neural Lyapunov Control for Nonlinear Systems with Unstructured Uncertainties

This GitHub repository contains the codebase for the paper Wei, S., Krishnamurthy, P., & Khorrami, F. (2023). Neural Lyapunov Control for Nonlinear Systems with Unstructured Uncertainties. arXiv preprint arXiv:2303.09678, which is also accepted at the 2023 American Control Conference (ACC).

Note: Part of the code is adapted from https://github.com/amehrjou/neural_lyapunov_redesign

Table of Contents:

1. Prerequisites
2. Code Explained
3. Code Usage

Prerequisites

The code is tested on

• Python 3.8.15
• PyTorch 1.13.0
• dReal4

Code Explained

The codebase contains three examples: the inverted pendulum (eg1_inverted_pendulum_2d), a hypothetical system of strict feedback form (eg2_backstepping_3d), and the cart-pole (eg3_cartpole_4d).

Code Usage

1. Open the main script, e.g. eg1_inverted_pendulum_2d/inv_pend_2d_sum4_nn_controller.py and modify the hyper-parameters in input_args_str located in the beginning of the script (after the imports)
2. Execute the script
3. After training, the results will be saved in the results/exp_XXX where XXX is the exp_num you defined in the hyper-parameters. The script will automatically save all the hyper-parameters in results/exp_XXX/00hyper_parameters.txt.
4. The scripts diagnostic*.py and post_processing*.py are used for analyzing and visualizing the results.