We propose a new framework for Imitation Learning (IL) via density estimation of the expert's occupancy measure followed by Maximum Occupancy Entropy Reinforcement Learning (RL) using the density as a reward. Our approach maximizes a non-adversarial model-free RL objective that provably lower bounds reverse Kullback-Leibler divergence between occupancy measures of the expert and imitator. We present a practical IL algorithm, Neural Density Imitation (NDI), which obtains state-of-the-art demonstration efficiency on benchmark control tasks.
@article{arxiv.2010.09808,
title = {Imitation with Neural Density Models},
author = {Kuno Kim and Akshat Jindal and Yang Song and Jiaming Song and Yanan Sui and Stefano Ermon},
journal= {arXiv preprint arXiv:2010.09808},
year = {2020}
}