Imitation Learning for compressing MPC.

This repo contains an implementation of using imitation learning to copy the behavior of an mpc.
To make the implementation easier (bypass version conflicts and etc), we borrowed several utility functions from:
Stable Baseline3 and Tao Chen's repo
The rl finetuning code is from: CleanRL.

Installation

1. Create a python virtual environment (I used python 3.10). The following steps are all done inside the venv.
2. Clone Tao Chen's repo and run python setup.py install inside that repo.
3. Install pytorch, scipy, pwlf via pip install.
4. Then you can work through the example in bc_trainer to get an idea of the workflow.

The file double_integrator includes the expert mpc, and some sample data were collected and stored in data_multimodal.

To start, the main file used for behavior cloning is the bc_trainer file, which parses the collected data and trains a student policy based on the processed data. After installing the requirements listed at the begging of the notebook, users should be able to run the code in the notebook to generate a student policy as well as to test it.

The radar_maps folder contains a gym environment that simulates the tracking problem.