An Approach to Partial Observability in Games: Learning to Both Act and Observe

TL;DR

This paper introduces a method for reinforcement learning agents to learn both where to look and how to act in partially observable environments, demonstrated through Atari and Pong games.

Contribution

It develops a neural network architecture and training approach enabling agents to learn visual attention strategies alongside gameplay actions in limited visibility scenarios.

Findings

Agents learn effective visual attention strategies.

The method improves performance in partially observable environments.

Analysis reveals how agents develop their attention and action policies.

Abstract

Reinforcement learning (RL) is successful at learning to play games where the entire environment is visible. However, RL approaches are challenged in complex games like Starcraft II and in real-world environments where the entire environment is not visible. In these more complex games with more limited visual information, agents must choose where to look and how to optimally use their limited visual information in order to succeed at the game. We verify that with a relatively simple model the agent can learn where to look in scenarios with a limited visual bandwidth. We develop a method for masking part of the environment in Atari games to force the RL agent to learn both where to look and how to play the game in order to study where the RL agent learns to look. In addition, we develop a neural network architecture and method for allowing the agent to choose where to look and what action to take in the Pong game. Further, we analyze the strategies the agent learns to better understand how the RL agent learns to play the game.