cs.CVNov 11, 2026

Prediction or Persistence? What Future-Prediction Scores Miss About World Models

Authors: Animesh Maheshwari

Organizations: Deccan AI

Abstract

Pretrained video world models are compared and chosen as components for robot learning largely by future-prediction scores. On its own, such a score cannot be read as evidence of prediction, because it mixes encoder quality, the near future's resemblance to the recent past, and the predictor's own contribution. We separate the three with a matched control. On 27,000 future-retrieval trials from household egocentric video, a model must pick the true continuation of a clip among eleven candidates from the same video (chance 0.091), ranked zero-shot by its own training criterion. Each model is paired with a persistence control, an identical copy whose predictor is never invoked, so the paired difference isolates the predictor. Score and contribution dissociate within the native-video V-JEPA family alone: the predictor adds +0.291 top-1 for V-JEPA 2.1 but +0.03 to +0.04 for V-JEPA 2 on the identical trials. Across five models the contribution ranges from +0.291 to negative, and it is negative only for the two action-conditioned models, which we evaluate outside their training regime. Persistence alone recovers the encoder ordering expected from pretraining, a sign that the probe tracks real signal. We suggest reporting each future-prediction score together with its persistence baseline. The gap between them, not the score, is what makes the number interpretable: a necessary check costing one forward pass. Establishing downstream usefulness is a separate matter, for a planning or control evaluation.

Figures & tables

Appendix figures & tables15 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Oct 5, 2026

Tracking Is Not Permanence: What Video World Models Keep of a Hidden Object

Video world models track objects they can see; we ask what they keep of objects they cannot. We hide an object from a frozen V-JEPA 2 predictor and compare its prediction for the hidden region with the encoder's representation of two worlds that differ only inside that region. The predictor's decision keeps a stationary object in part and one carried inside a container not at all, and loses a moving one within 0.3 s (0.5 s under V-JEPA's own tube mask; ViT-H keeps it to 1.1 s at pretraining's 90% masking ratio); in projection a trace remains, below the midpoint, at 14-60% of what a baseline copying the last view retains. The information is there: the encoder reads the object's presence at 1.00 and keeps a closed container's contents decodable for 3.5 s, while the predictor's output, read with the encoder's own probe, contains the ball in 2% of scenes once the box has been closed for half a second. On rendered scenes, permanence is missing on the predictor's side, and training installs it cheaply as a prior: three thousand predictor-only steps on synthetic containers take this belief from 0.05 to 1.00 against two matched controls. They also raise IntPhys-2019 from 84.2% to 93.3%, but so does a curriculum without containers, and which training habit the benchmark credits changes with its scoring rule. Continued training with tube masks produces 1.1-1.6 s of moving-object carry-over on manipulation and internet-style video, so the deficit is not intrinsic to latent prediction. VideoMAE keeps almost nothing, and Cosmos's next-token prediction keeps a stationary hidden object but not one carried inside a moving container.
Nov 15, 2026

The Flip Test: Video World Models Fail to Learn the World State

Video world models promise to predict the future of a scene from its present. Do current models keep this promise? We probe this question counterfactually. When a small change to the present flips the future, such as a box resting a few millimeters before or past a table edge, a model that tracks the state of the world must flip its prediction too. We build FlipTest, a three-level hierarchy of increasingly strict tests over 1,680 initial frames from 10 simulated and 15 real-world tasks. At the base, visual minimal pairs, two nearly identical frames whose true futures differ, test whether predictions change across the boundary; matched controls go further and test whether they change only where the future does; and dense sweeps, the strictest level, locate exactly where they change. All 11 video world models we evaluate fall short. The best averages 18.3% pair accuracy over real and simulated scenes, below the 25% of random guessing. Models flip as often on steps that leave the future unchanged as on steps that change it, and their response stays flat across the sweeps. On the same frames, a vision-language model is selective and traces a near S-shaped curve near the boundary, and a coding agent that measures the image comes close to the true boundary. This implies that the deciding information is in the frame, and video models just miss it. We therefore urge the community to rethink world modeling as pure video prediction. Next-frame prediction must do more than draw plausible pixels; it must also identify the causal states that control future dynamics.
Oct 1, 2026

FutureWorlds: Learning Robotic World Models from Alternative Futures

Robotic world models predict action-conditioned future scenes, providing a foundation for understanding action outcomes. However, turning alternative predictions into useful learning signals remains challenging: similar candidates limit informative quality comparisons, while diverging trajectories require persistent maintenance of their individual histories. We introduce FutureWorlds, a framework that unifies candidate construction, history maintenance, and learning from relative quality. Built on a multimodal discrete autoregressive model, FutureWorlds uses diverse beam search during reinforcement learning to construct candidate futures that balance confidence and diversity. Candidate-specific bounded memory preserves scene states and ensures that generation and policy scoring use matching histories. We further propose MemSPO (Memory-Conditioned Search-Guided Policy Optimization), which converts video trajectory rewards into group-relative advantages to optimize the world model. On RT-1, BridgeV2, and RoboCasa, FutureWorlds reduces LPIPS for 32-frame predictions by 14.78%, 20.84%, and 9.12%, respectively, relative to the strongest baseline on each dataset. Under fixed evaluation configurations, only 200 MemSPO updates further improve generation quality and support continued prediction beyond the training horizon. Memory ablations, decoding sensitivity analysis, and optical-flow evaluation show that these gains extend beyond visual quality to more accurate motion prediction and more consistent object states. Project page and code: https://github.com/Alexander-wu/FutureWorlds.