Visual Hide and Seek

TL;DR

This paper presents a study on training embodied agents to play a Visual Hide and Seek game, where agents learn to hide from predators in a simulated environment using partial observations, revealing insights into learned visibility prediction and the impact of agent weaknesses.

Contribution

It introduces a novel environment and training setup for visual hide and seek, demonstrating that agents learn to predict visibility and that weaknesses influence feature learning.

Findings

Agents learn to predict their own visibility.

Agent weaknesses lead to more useful feature learning.

Slower agents face more challenging learning but develop better features.

Abstract

We train embodied agents to play Visual Hide and Seek where a prey must navigate in a simulated environment in order to avoid capture from a predator. We place a variety of obstacles in the environment for the prey to hide behind, and we only give the agents partial observations of their environment using an egocentric perspective. Although we train the model to play this game from scratch, experiments and visualizations suggest that the agent learns to predict its own visibility in the environment. Furthermore, we quantitatively analyze how agent weaknesses, such as slower speed, effect the learned policy. Our results suggest that, although agent weaknesses make the learning problem more challenging, they also cause more useful features to be learned. Our project website is available at: http://www.cs.columbia.edu/ ~bchen/visualhideseek/.