AI summaryⓘ
The authors explain that what we now call 'world models' — simplified internal maps of environments used for predicting and planning — were actually developed decades earlier in engineering fields under different names. They trace how three groups contributed parts: turbulence modeling gave data-driven latent dynamics, early computer vision provided encoding and decoding methods, and thermal control systems combined these into a full loop with built-in checks to ensure predictions were reliable. The authors highlight that while modern world models add nonlinearity and adaptability, older methods offer strong verification and physical understanding. They suggest future work should merge these to build trustworthy models for critical systems like power or process control.
world modelslatent representationsmodel order reductionproper orthogonal decompositionencoder-decoderverificationcontrol systemsself-supervised learningnonlinear representationpredictive fidelity
Abstract
World models -- compressed latent representations of an environment that support action-conditioned prediction and planning -- are typically presented as a product of modern self-supervised learning. This paper argues that the functional anatomy of a world model was independently developed, deployed, and formally analyzed decades earlier in the model-order-reduction (MOR) and control literature, under different names and for a different purpose: the real-time operation of physical systems. We trace the anatomy across three communities. Low-dimensional models of turbulence built on proper orthogonal decomposition (POD) supplied latent dynamics learned from data of a chaotic environment; eigenface methods in early computer vision supplied the encoder-decoder half, including a primitive runtime validity check; and measurement-based POD frameworks for facility thermal control assembled the complete loop -- POD coefficients as latent state, parametric dependence on actuator setpoints as action conditioning, modal reconstruction as decoding, and, critically, a priori analytical error bounds as a verification layer that certified when the model's predictions could be trusted in closed loop. We then examine what each tradition possesses that the other lacks: MOR contributes verification, physical grounding, and extreme data efficiency; learned world models contribute nonlinear representation, transferability, and horizon. We argue that the outstanding obstacle to deploying world models in systems that cannot fail -- power, thermal, process control -- is not predictive fidelity but verifiability, and we outline a research agenda for physics-grounded, verifiable world models that unifies the two lineages.