Abstract
Reinforcement learning (RL) has proven its worth in a series of artificialdomains, and is beginning to show some successes in real-world scenarios.However, much of the research advances in RL are hard to leverage in real-worldsystems due to a series of assumptions that are rarely satisfied in practice.In this work, we identify and formalize a series of independent challenges thatembody the difficulties that must be addressed for RL to be commonly deployedin real-world systems. For each challenge, we define it formally in the contextof a Markov Decision Process, analyze the effects of the challenge onstate-of-the-art learning algorithms, and present some existing attempts attackling it. We believe that an approach that addresses our set of proposedchallenges would be readily deployable in a large number of real worldproblems. Our proposed challenges are implemented in a suite of continuouscontrol environments called realworldrl-suite which we propose an as anopen-source benchmark.