LiRA: Light-Robust Adversary for Model-based Reinforcement Learning

Taisuke Kobayashi

Keywords: Reinforcement learning, Adversarial learning, Variational inference, Light robustness

TL;DR: This paper proposes a moderately-robust adversarial learning methodology

Abstract:

This study addresses a new adversarial learning framework to make reinforcement learning robust moderately and not conservative too much.
To this end, the adversarial learning is first rederived with variational inference.
In addition, $\textit{light robustness}$, which allows for maximizing robustness within an acceptable performance degradation, is utilized as a constraint.
As a result, the proposed framework, so-called LiRA, can automatically adjust adversary level, balancing robustness and conservativeness.
The behaviors of LiRA are confirmed in numerical simulations.

Submission Number: 15