Investigating Memory in Model-Free RL with POPGym Arcade
For researchers studying memory in deep RL, this work provides tools and reveals a specific failure mode, though the findings are incremental.
The paper introduces POPGym Arcade, a set of Atari-inspired environments with fully and partially observable variants, and uses it to analyze memory in deep RL. It identifies a pathology where value functions smear credit over irrelevant history, showing how out-of-distribution scenarios can contaminate memory and perturb policies far into the future.
How should we analyze memory in deep RL? We introduce tools for analyzing policies under partial observability and revealing how agents use memory to make decisions. To utilize these tools, we present POPGym Arcade, a collection of Atari-inspired, hardware-accelerated environments sharing a single observation and action space. Each environment provides fully and partially observable variants, enabling counterfactual studies on observability. We find that controlled studies are necessary for fair comparisons and identify a pathology where value functions smear credit over irrelevant history. Using this pathology, we demonstrate how out-of-distribution scenarios can contaminate memory, perturbing the policy far into the future. Our code is available at https://github.com/bolt-research/popgym-arcade.