Proceedings of the Annual Meeting of the Cognitive Science Society

We explore the benefits of augmenting state-of-the-art model-free deep reinforcement learning with simple object representa-tions. Following the Frostbite challenge posited by Lake et al.(2017), we identify object representations as a critical cognitivecapacity lacking from current reinforcement learning agents.We discover that providing the Rainbow model (Hessel et al.,2018) with simple, feature-engineered object representationssubstantially boosts its performance on the Frostbite game fromAtari 2600. We then analyze the relative contributions of therepresentations of different types of objects, identify environ-ment states where these representations are most impactful, andexamine how these representations aid in generalizing to novelsituations.