Towards Deployable RL -- What's Broken with RL Research and a Potential Fix

TL;DR

This paper critiques current reinforcement learning research for overhyping and practical limitations, and proposes a potential fix to make RL more deployable in real-world, economically viable scenarios.

Contribution

It identifies key issues hindering RL deployment and suggests a possible approach to address these challenges, aiming to bridge the gap between research and practical application.

Findings

Current RL research is overly hyped and impractical.

Identifies endemic issues in RL research direction.

Proposes a potential fix to improve RL deployability.

Abstract

Reinforcement learning (RL) has demonstrated great potential, but is currently full of overhyping and pipe dreams. We point to some difficulties with current research which we feel are endemic to the direction taken by the community. To us, the current direction is not likely to lead to "deployable" RL: RL that works in practice and can work in practical situations yet still is economically viable. We also propose a potential fix to some of the difficulties of the field.