Relational Deep Reinforcement Learning
This addresses stubborn challenges in deep RL for applications requiring complex planning and generalization, representing a novel method rather than an incremental improvement.
The paper tackles the problem of improving efficiency, generalization, and interpretability in deep reinforcement learning by introducing a method that uses self-attention for relational reasoning between entities. The results show state-of-the-art performance on six StarCraft II mini-games (surpassing human grandmaster on four) and improved sample complexity and generalization in a Box-World task.
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.