Leveraging class abstraction for commonsense reinforcement learning via residual policy gradient methods
This addresses the challenge of integrating unstructured knowledge into RL for knowledge-intensive domains, though it is incremental as it builds on existing methods with specific adaptations.
The paper tackled the problem of enabling reinforcement learning agents to leverage open-source knowledge graphs without manual tailoring, by using class abstraction and residual policy gradient methods, resulting in improved sample efficiency and generalization to unseen objects in commonsense games.
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.