Challenges of Real-World Reinforcement Learning

Gabriel Dulac-Arnold, Daniel Mankowitz, Todd Hester

Keywords: reinforcement learning, real-world, applications, benchmarks

Abstract: Reinforcement learning (RL) has proven its worthin a series of artificial domains, and is beginningto show some successes in real-world scenarios.However, much of the research advances in RLare often hard to leverage in real-world systemsdue to a series of assumptions that are rarely sat-isfied in practice. We present a set of nine uniquechallenges that must be addressed to production-ize RL to real world problems. For each of thesechallenges, we specify the exact meaning of thechallenge, present some approaches from the liter-ature, and specify some metrics for evaluating thatchallenge. An approach that addresses all ninechallenges would be applicable to a large numberof real world problems. We also present an exam-ple domain that has been modified to present thesechallenges as a testbed for practical RL research.

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:1904.12901/code)