When in Doubt, Think Slow: Iterative Reasoning with Latent Imagination
This work addresses performance limitations for reinforcement learning agents in partially-observable environments, offering an incremental improvement over existing methods.
The paper tackles the problem of model-based reinforcement learning agents being limited by inaccurate world models in unfamiliar settings, and presents a training-free iterative inference method that improves reconstruction accuracy and task performance in visual 3D navigation tasks, with agents benefiting most when less trained.
In an unfamiliar setting, a model-based reinforcement learning agent can be limited by the accuracy of its world model. In this work, we present a novel, training-free approach to improving the performance of such agents separately from planning and learning. We do so by applying iterative inference at decision-time, to fine-tune the inferred agent states based on the coherence of future state representations. Our approach achieves a consistent improvement in both reconstruction accuracy and task performance when applied to visual 3D navigation tasks. We go on to show that considering more future states further improves the performance of the agent in partially-observable environments, but not in a fully-observable one. Finally, we demonstrate that agents with less training pre-evaluation benefit most from our approach.