Abstract
An embodied system must not only model the patterns of the external world butalso understand its own motion dynamics. A motion dynamic model is essentialfor efficient skill acquisition and effective planning. In this work, weintroduce the neural motion simulator (MoSim), a world model that predicts thefuture physical state of an embodied system based on current observations andactions. MoSim achieves state-of-the-art performance in physical stateprediction and provides competitive performance across a range of downstreamtasks. This works shows that when a world model is accurate enough and performsprecise long-horizon predictions, it can facilitate efficient skill acquisitionin imagined worlds and even enable zero-shot reinforcement learning.Furthermore, MoSim can transform any model-free reinforcement learning (RL)algorithm into a model-based approach, effectively decoupling physicalenvironment modeling from RL algorithm development. This separation allows forindependent advancements in RL algorithms and world modeling, significantlyimproving sample efficiency and enhancing generalization capabilities. Ourfindings highlight that world models for motion dynamics is a promisingdirection for developing more versatile and capable embodied systems.