Playing Games in the Dark: An approach for cross-modality transfer in reinforcement learning
This addresses the challenge of cross-modality transfer in reinforcement learning for agents operating in multimodal settings, though it appears incremental as it builds on existing methods.
The paper tackles the problem of enabling reinforcement learning agents to transfer policies across different sensory modalities, such as from images to sounds, by proposing a three-stage architecture. The result shows that these generalized policies achieve better out-of-the-box performance compared to baselines in various environments.
In this work we explore the use of latent representations obtained from multiple input sensory modalities (such as images or sounds) in allowing an agent to learn and exploit policies over different subsets of input modalities. We propose a three-stage architecture that allows a reinforcement learning agent trained over a given sensory modality, to execute its task on a different sensory modality-for example, learning a visual policy over image inputs, and then execute such policy when only sound inputs are available. We show that the generalized policies achieve better out-of-the-box performance when compared to different baselines. Moreover, we show this holds in different OpenAI gym and video game environments, even when using different multimodal generative models and reinforcement learning algorithms.