RTFM: Generalising to Novel Environment Dynamics via Reading
This addresses the problem of generalization in reinforcement learning for AI agents, offering a novel approach through language understanding, though it is incremental in combining reading with policy learning.
The paper tackles the challenge of generalizing reinforcement learning policies to new environments by introducing a grounded policy learning problem called RTFM, where agents must read language descriptions to understand novel dynamics, and proposes txt2π, a model that outperforms baselines like FiLM and language-conditioned CNNs on this task.
Obtaining policies that can generalise to new environments in reinforcement learning is challenging. In this work, we demonstrate that language understanding via a reading policy learner is a promising vehicle for generalisation to new environments. We propose a grounded policy learning problem, Read to Fight Monsters (RTFM), in which the agent must jointly reason over a language goal, relevant dynamics described in a document, and environment observations. We procedurally generate environment dynamics and corresponding language descriptions of the dynamics, such that agents must read to understand new environment dynamics instead of memorising any particular information. In addition, we propose txt2$π$, a model that captures three-way interactions between the goal, document, and observations. On RTFM, txt2$π$ generalises to new environments with dynamics not seen during training via reading. Furthermore, our model outperforms baselines such as FiLM and language-conditioned CNNs on RTFM. Through curriculum learning, txt2$π$ produces policies that excel on complex RTFM tasks requiring several reasoning and coreference steps.