Deep Recurrent Q-Learning for Partially Observable MDPs

TL;DR

This paper introduces a Deep Recurrent Q-Network (DRQN) that incorporates recurrency into Deep Q-Learning, enabling better handling of partial observations in complex tasks like Atari games by integrating information over time.

Contribution

The paper demonstrates that adding recurrency to DQN allows effective memory and adaptation to partial observations, matching performance on standard tasks and improving robustness to observation quality changes.

Findings

DRQN performs comparably to DQN on standard Atari games.

DRQN better handles partial and flickering observations.

Recurrency enhances adaptability to changing observation quality.

Abstract

Deep Reinforcement Learning has yielded proficient controllers for complex tasks. However, these controllers have limited memory and rely on being able to perceive the complete game screen at each decision point. To address these shortcomings, this article investigates the effects of adding recurrency to a Deep Q-Network (DQN) by replacing the first post-convolutional fully-connected layer with a recurrent LSTM. The resulting \textit{Deep Recurrent Q-Network} (DRQN), although capable of seeing only a single frame at each timestep, successfully integrates information through time and replicates DQN's performance on standard Atari games and partially observed equivalents featuring flickering game screens. Additionally, when trained with partial observations and evaluated with incrementally more complete observations, DRQN's performance scales as a function of observability. Conversely, when trained with full observations and evaluated with partial observations, DRQN's performance degrades less than DQN's. Thus, given the same length of history, recurrency is a viable alternative to stacking a history of frames in the DQN's input layer and while recurrency confers no systematic advantage when learning to play the game, the recurrent net can better adapt at evaluation time if the quality of observations changes.