We have implemented two versions of FLAIR, based on two difference libraries - rllab and Garage.
We did our experiment on the Inverted Pendulum with the rllab implementation and the Lunar Lander and Bipedal Walker with the Garage implementation. In general, the garage-based implementation is more advanced as it supports parallelization for environment rollouts using Ray.
Detailed READMEs for how to use the implementations are inside their folders.
If you use this software please cite as follows:
@inproceedings{
chen2022flair,
title={Fast Lifelong Adaptive Inverse Reinforcement Learning from Demonstrations},
author={Letian Chen and Sravan Jayanthi and Rohan R Paleja and Daniel Martin and Viacheslav Zakharov and Matthew Gombolay},
booktitle={Proceedings of Conference on Robot Learning (CoRL)},
year={2022}
}