Relational Reinforcement Learning in Infinite Mario

TL;DR

This paper demonstrates how relational representations in reinforcement learning enable the incorporation of background knowledge, facilitating learning in complex domains like large-scale computer games.

Contribution

It introduces the use of relational representations in reinforcement learning for large state and action spaces, enhancing the ability to include background knowledge.

Findings

Relational representations improve learning efficiency in complex environments.

Background knowledge integration enhances agent performance.

Applicable to large-scale computer game domains.

Abstract

Relational representations in reinforcement learning allow for the use of structural information like the presence of objects and relationships between them in the description of value functions. Through this paper, we show that such representations allow for the inclusion of background knowledge that qualitatively describes a state and can be used to design agents that demonstrate learning behavior in domains with large state and actions spaces such as computer games.