Current World Models Lack a Persistent State Core

2026-06-18

AI Summary

This work addresses a critical limitation in existing world models: their inability to maintain coherent internal state evolution when targets are occluded, leading to inconsistent event representations upon viewpoint changes. To systematically evaluate this issue, the authors introduce WRBench, a benchmark that treats camera motion as an observability intervention and assesses model performance in state continuity and event consistency when objects reappear. The benchmark innovatively incorporates a human-calibrated chain-of-evaluation protocol that explicitly disentangles rendering quality from state persistence, alongside controllable camera trajectories, scene identifiability analysis, and event consistency verification. Experiments across 23 state-of-the-art models and 9,600 videos reveal that none can reliably advance internal states during target occlusion, underscoring persistent world state modeling as an unresolved challenge in artificial intelligence.

Abstract

World models are increasingly regarded as a decisive step toward artificial general intelligence, yet modeling the physical world demands more than rendering convincing frames on demand: it requires an internal world state that keeps evolving over time, decoupled from observation, so that objects endure and events run to their conclusions whether or not a camera is watching, much as the moon holds to its orbit when no one is looking. This requirement is a blind spot of existing benchmarks, which reward surface properties such as fidelity, motion, and camera controllability while never asking whether a generated world keeps evolving once it is unobserved. We introduce \textbf{WRBench}, the first systematic diagnostic benchmark that treats camera motion as an intervention on observability and resolves evaluation into a human-calibrated chain that asks whether the camera executes the requested interaction, whether the scene stays continuous and identifiable while in view, and whether a returning target remains consistent with the event that was set in motion. Across 9{,}600 videos from 23 models spanning four control paradigms, one finding proves stubborn: current systems maintain the observed world as a tracking shot, resuming a returning target in the state at which it was abandoned rather than advancing the event while it went unseen. Because this failure recurs across control paradigms, model families, and increments of scale, robust world-state evolution does not follow from cleaner imagery, tighter control, richer geometric priors, or sheer parameter count We therefore argue that the stability of the physical state kernel and the consistency of worldlines under viewpoint intervention should become first-class objectives of world-model design, so that a world model captures how the world will unfold rather than how the next frame appears.

Problem

• world models
• persistent state
• unobserved evolution
• world consistency
• state continuity

Innovation

• persistent state core
• world models
• WRBench
• state evolution
• observability intervention