Relational Deep Reinforcement Learning

TL;DR

This paper presents a relational deep reinforcement learning approach that enhances efficiency, generalization, and interpretability by using self-attention for relational reasoning, demonstrating superior performance in navigation, planning, and StarCraft II mini-games.

Contribution

Introduces a relational deep RL method utilizing self-attention for structured reasoning, improving sample efficiency, generalization, and interpretability over conventional methods.

Findings

Outperforms baselines in the Box-World navigation task.

Achieves state-of-the-art results on six StarCraft II mini-games.

Surpasses human grandmaster performance on four StarCraft II mini-games.

Abstract

We introduce an approach for deep reinforcement learning (RL) that improves upon the efficiency, generalization capacity, and interpretability of conventional approaches through structured perception and relational reasoning. It uses self-attention to iteratively reason about the relations between entities in a scene and to guide a model-free policy. Our results show that in a novel navigation and planning task called Box-World, our agent finds interpretable solutions that improve upon baselines in terms of sample complexity, ability to generalize to more complex scenes than experienced during training, and overall performance. In the StarCraft II Learning Environment, our agent achieves state-of-the-art performance on six mini-games -- surpassing human grandmaster performance on four. By considering architectural inductive biases, our work opens new directions for overcoming important, but stubborn, challenges in deep RL.