World Models as Reference Trajectories for Rapid Motor Adaptation

TL;DR

This paper introduces Reflexive World Models, a dual control framework that enables rapid adaptation in real-world control tasks by using world model predictions as implicit reference trajectories, combining reinforcement learning with quick error correction.

Contribution

The paper presents a novel dual control architecture that separates reward maximization from rapid motor adaptation, significantly improving adaptation speed and computational efficiency.

Findings

Faster adaptation compared to model-based RL baselines

Maintains near-optimal performance under dynamic changes

Low online computational cost

Abstract

Deploying learned control policies in real-world environments poses a fundamental challenge. When system dynamics change unexpectedly, performance degrades until models are retrained on new data. We introduce Reflexive World Models (RWM), a dual control framework that uses world model predictions as implicit reference trajectories for rapid adaptation. Our method separates the control problem into long-term reward maximization through reinforcement learning and robust motor execution through rapid latent control. This dual architecture achieves significantly faster adaptation with low online computational cost compared to model-based RL baselines, while maintaining near-optimal performance. The approach combines the benefits of flexible policy learning through reinforcement learning with rapid error correction capabilities, providing a principled approach to maintaining performance in high-dimensional continuous control tasks under varying dynamics.