Leveraging class abstraction for commonsense reinforcement learning via residual policy gradient methods

TL;DR

This paper introduces a residual policy gradient method that leverages class hierarchies from knowledge graphs to improve reinforcement learning in knowledge-rich environments, enhancing sample efficiency and generalization.

Contribution

It presents a novel approach to incorporate class abstraction from knowledge graphs into RL via residual policy gradients, addressing challenges in leveraging unstructured knowledge.

Findings

Improved sample efficiency in commonsense games.

Enhanced generalization to unseen objects.

Identified limitations due to noisy class knowledge.

Abstract

Enabling reinforcement learning (RL) agents to leverage a knowledge base while learning from experience promises to advance RL in knowledge intensive domains. However, it has proven difficult to leverage knowledge that is not manually tailored to the environment. We propose to use the subclass relationships present in open-source knowledge graphs to abstract away from specific objects. We develop a residual policy gradient method that is able to integrate knowledge across different abstraction levels in the class hierarchy. Our method results in improved sample efficiency and generalisation to unseen objects in commonsense games, but we also investigate failure modes, such as excessive noise in the extracted class knowledge or environments with little class structure.