World Model Evaluation

Follow this topic Ask about this topic

Momentum

29 papers in the last four weeks, up 123% on the four weeks before. 0.3% of all new papers.

Jul 13Week of Sep 28

Latest papers 146

Jul 20, 2026cs.AI

AutoWorldModel-Bench: A State-Centric Benchmark for Automated World-Model Research

World modeling is an unsettled field: architectures, training objectives, and state representations interact in complex ways, and no single recipe dominates across environments. This makes it an ideal testbed for AI coding agents acting as autonomous researchers--a setting in which the improvement direction is not specified in advance, unlike the engineering-to-spec tasks that dominate current agent benchmarks. We introduce AutoWorldModel-Bench, a closed-loop benchmark in which frontier coding agents autonomously improve a provided world-model starter under a fixed compute budget. The benchmark spans eight game environments under a unified structured-state representation--ground-truth entity state extracted from each game and consumed through a shared tensor format--which isolates dynamics modeling from perception and enables minutes-per-run iteration. Across 64 sessions, Codex-5.4 and Claude Opus 4.6 improve their starter on 63; in 91% of sessions the winning edit is a non-trivial research-style modification--a new objective, representation, rollout procedure, or architectural change--rather than a hyperparameter tweak. Our benchmark offers a setting in which frontier coding agents can be evaluated on open-ended research rather than engineering-to-spec problems.

Marjan Moodi, Xuankang Zhu, Fernando De Mesentier Silva +2

Electronic Arts · Simon Fraser University

World Model Learning AI Agent Benchmarks World Model Evaluation Autonomous Research Agents

Jul 20, 2026cs.CV

Thinking in Video: Can Video Generators Really Reason About the Real World?

Recent advances in world models and video generation have given rise to an emerging reasoning paradigm that leverages video generative models to simulate, predict, and reason about real-world dynamics. We redefine this paradigm as Thinking in Video, where video is not merely an output artifact but a medium for constructing, extending, and verifying causal thought. However, this promise remains unverified: convincing rollouts may reflect memorized appearances rather than causal understanding, while existing metrics separate perceptual fidelity from semantic logic. To evaluate whether video generators support such reasoning, we introduce the Causal-Generative Dual-Judge (CGDJ), auditing World Model Consistency from two perspectives. Explicit Causal Perception tests whether a generator reads a video scenario as a reasoning problem through spatio-temporal flattened visual question answering, while Implicit Generative Perception-Prediction Gap evaluates whether it renders the causal consequence as a consistent future video. Applying CGDJ to representative open- and closed-source generators reveals a clear Perception-Prediction Gap: open-source models produce plausible dynamics despite near-zero explicit causal perception, whereas advanced closed-source systems show stronger but still limited alignment between reasoning and generation. Further analysis exposes audio-visual misalignment, where models verbalize correct causal logic more reliably than they render it, challenging the "world simulator" narrative.

Yongheng Zhang, Guang Yang, Ruihan Hou +12

Central South University · Tencent · Tsinghua University

Video Reasoning Causal Reasoning in Videos World Model Evaluation Video Generation Evaluation

Jul 16, 2026cs.AI

Concept-Guided Spatial Regularization for World Models in Atari Pong

World models are usually evaluated as components of model-based reinforcement learning (MBRL) systems, while the world models themselves are rarely studied in isolation. We examine five representative visual world-model agents in Atari Pong: DreamerV3, DIAMOND, TWISTER, Simulus, and STORM. After reproducing their training pipelines and matching the reported agent performance, we freeze the learned world models and evaluate them with a closed-loop rollout diagnostic: a policy trained separately from the corresponding MBRL agent interacts with each frozen model, and the generated video trajectories are inspected for visual and dynamical errors. Across all five models, the rollouts contain clear failures, including ball disappearance, incorrect ball motion, and invalid ball-paddle interactions. Beyond visual trajectories, we further evaluate them with pixel-space zero-shot MBRL, where a new policy is trained entirely inside a frozen world model and then evaluated in the real environment. Across all five models, the resulting policies substantially underperform those produced by the corresponding original MBRL training pipelines. The gap is particularly large for DreamerV3, whose mean return drops from -5.5 to -20.9, near the minimum Pong return of -21. We hypothesize that insufficient modeling of task-critical concepts, such as the ball in Pong, may contribute to these failures. We therefore propose Concept-Guided Spatial Regularization (CGSReg), an auxiliary pixel reconstruction loss applied to segmented concept regions. Experiments show that CGSReg improves both closed-loop rollouts and pixel-space zero-shot MBRL in DreamerV3, DIAMOND, and TWISTER. Its effects vary across the remaining models and evaluation metrics, indicating that CGSReg alone does not address all world-model bottlenecks.

Yukuan Lu, Zaishuo Xia, Weyl Lu +1

UC Davis

Game-Playing Agents World Model Learning World Model Evaluation Model-Based RL

Jul 15, 2026cs.CV

Towards Spatial Supersensing in the Wild

Humans can efficiently parse continuous sensory streams, from hours to years, scaffolding an internal world model that grounds spatial reasoning and prediction. To mimic this capacity, spatial supersensing challenges multimodal models to move beyond linguistic understanding toward true world modeling. However, their benchmark relies on synthetic long videos, formed by concatenating random short clips, and is mostly limited to household scenes, leaving real-world continuity and diversity underexplored. To address the gap, we introduce

\textbf{VSI-Super-Wild}

, a large-scale benchmark for evaluating spatial supersensing over long temporal horizons in diverse in-the-wild scenes. Notably, inspired by cognitive studies on how humans structure experience, we systematically probe the full triad of world state: the agent (observer), objects (scene items), and the environment (places and global layout). In total, VSI-Super-Wild contains

\textbf{6,980}

human-verified question-answer pairs derived from

\textbf{442}

real-world videos spanning 8 scene categories, including long-form recordings exceeding 4 hours. Results on VSI-Super-Wild expose a fundamental disconnect: despite advances in static image understanding, models consistently fail at tasks that require coherent world-state tracking over time. We characterize how performance degrades with world-state complexity and temporal horizon, and diagnose four failure modes: spatial collapse, semantic shortcuts, insufficient update, and instance confusion. This taxonomy reveals that models lack mechanisms to bind objects, agents, and environments into a unified spatial world model, a fundamental gap that defines the path forward for spatial supersensing.

Tianjun Gu, Tianyu Xin, Kuan Zhang +12

Tsinghua University · NVIDIA · Stanford University

Long-Video Understanding World Model Evaluation Temporal Video Understanding

Jul 15, 2026cs.AI

When a Verified World Model Still Loses: Play-Adequacy vs Prediction-Accuracy in LLM-Synthesized Code World Models

Large language models can synthesize a game's rules as executable code - a Code World Model (CWM) - which a classical planner then searches over. Such models are typically accepted when they reach high transition accuracy on sampled trajectories. We argue this is the wrong notion of adequacy for planning. We show four things. (1) An LLM-synthesized CWM can pass a sampling gate at 100% transition accuracy and be

\geq 98\%

state-accurate on the planner's own search distribution, yet lose systematically at play, because the

<1\%

it gets wrong is exactly the pivotal dynamics; the play cost of the omitted rule is

0.091

(seed-clustered 95% CI

[0.065,0.117]

,

n=4800

). We call this the verified-vs-correct gap, and confirm it end-to-end through the synthesis pipeline. (2) The harm follows a quantitative law,

\mathrm{danger}=\mathrm{play\_cost}\times(1-\mathrm{rarity})^N

, whose

(1-\mathrm{rarity})^N

gate-miss factor is proven exact and whose play cost is empirically bounded. (3) The failure is not repaired by more data: LLM synthesis behaves as rule translation, not rule inference, and did not infer the omitted rule across models (GPT-5.x) and data regimes (including DAgger and targeted examples). (4) The same mechanism recurs on the belief-inference function of imperfect-information CWMs: we prove a coverage bound (a size-

N

gate is identifying when

N\gtrsim b^{d_{\max}}

), explaining why shallow games such as Kuhn poker show no gap, and hand-construct Beacon, a verified-but-wrong inference function that passes the gate yet loses every game. These results suggest adequacy for planning-oriented world models should be measured on the search distribution or by play directly, not by prediction accuracy on sampled transitions.

Javier Aguilar Martín

AGILabs (javieraguilar.ai)

LLM World Models World Models World Model-Based Planning LLM-Based Program Synthesis Model-Based Planning World Model Evaluation

Jul 11, 2026cs.LG

A Control Theory of Predictability in Latent World Models

Latent world models are trained to predict future states in a learned representation and are then deployed inside a planner that selects actions by simulating them forward. Current practice adopts the prediction error, the single- or multi-step rollout loss on held-out data, as the training and model-selection objective, on the assumption that a lower prediction error yields better control. We show that this assumption is unreliable for a structural reason: a planner does not query the model on the training distribution but on the states that its candidate actions reach, which generally leave the data manifold, so an error averaged over the data cannot by itself govern control. We therefore reframe the objective as the discrepancy between the predicted and the true plan-cost at the plan the planner commits to, and prove that the planner's suboptimality is bounded by twice this discrepancy, whereas the data-averaged prediction error neither bounds nor tracks it. Under a linear-control premise the discrepancy separates into two terms. The first is a small on-manifold residual, on which the predicted and true dynamics agree and which a spectral tax prices through the non-normality of the latent transition operator. The second is an off-manifold divergence, on which an action carries the state off the manifold and the two dynamics diverge; this divergence is the binding term and is bounded by no data-averaged error. Synthetic operators confirm the pricing formulas, and latent model-predictive control experiments confirm the decoupling: across seeds, the single-step validation error is essentially uncorrelated with control success, whereas a fidelity score on the planner-reachable measure tracks it.

Hanzhe You, Yonggang Zhang, Maohao Ran +6

University of Science and Technology of China · The Hong Kong University of Science and Technology, HKGAI · Hong Kong Baptist University

Model Predictive Control Latent World Models World Model Evaluation

Jul 11, 2026cs.LG

Adaptive Compute in Latent World Models: When Depth Helps, Hurts, or Doesn't Matter

Adaptive-compute world models -- early-exit or mixture-of-depths predictors that spend variable depth per step -- assume depth buys better predictions and can be routed adaptively. In autoregressive rollouts, the first assumption requires depth's per-step precision to survive composition. We test this with a pre-registered instrument, the shallow penalty

ρ=\mathrm{err}(\text{shallowest-exit rollout})/\mathrm{err}(\text{full-depth rollout})

, across nine DeepMind Control tasks under matched single-step (

K=1

) and multi-step (

K=4

) training, three seeds each. We find three regimes: on 6/9 tasks depth helps rollouts (intrinsic,

ρ

up to

4.7\times

), on 2/9 the shallow exits beat the full stack (inversion,

ρ

down to

0.85\times

), and one is flat. The robust inversion (cheetah) is not a property of the dynamics but is created by training: an ablation supervising early exits only at the first rollout step erases it (

ρ: 0.87\to1.18

,

n=8

,

Δ=+0.31

), while an intrinsic-tradeoff task is unaffected -- a double dissociation we call the routability catch-22, since the supervision that makes exits routable is what trains them to out-roll the full stack. The regime is partly predictable a priori: observation/action dimensionality and one-step model error correlate with

ρ

at

|\text{Spearman}|\approx0.75

(

n=9

). Inside a CEM planner,

ρ

's sign predicts whether planning benefits from depth, most sharply on the inversion task, where shallow planning beats deep. Finally, three cautions: a task's regime depends on the metric space, the rollout horizon, and the encoder. All thresholds and gates were fixed before the compute campaign, including a pre-registered negative for the hypothesis that motivated the study.

Achyuthan Sivasankar

New York University

Latent World Models World Model Evaluation

Jul 8, 2026cs.RO

Validate the Dream Before You Trust Its Verdict: Admissibility for World-Model Simulators

Across robotics, World Models (WMs) are increasingly used to evaluate action policies by simulating the consequences of actions in an imagined world, and returning a success or safety verdict. Yet a verdict is only as trustworthy as the WM that produced it, and the WM itself needs to be certified. In video-generation WMs, fidelity metrics such as Fréchet Video Distance (FVD) reward visual realism, but ignore whether the world responds correctly to the policy's actions, including those unseen in training. Classical simulation-based validation assumes a trusted simulator evaluating an untrusted policy, whereas generative WMs are themselves unverified learned artifacts. Hence, we argue that any WM used as a test oracle must first be accredited before its verdicts can serve as evidence. Building on credibility practices from safety-critical simulation, including Verification, Validation & Accreditation (VV&A), Safety of the Intended Functionality (SOTIF), and scenario-based testing standards, we define an admissibility ladder (L0-L4) that a WM must climb before its closed-loop verdicts are accepted as assurance evidence. Our framework is embodiment-agnostic, and is instantiated in autonomous driving (AD), where assurance methods for traditional simulation are most mature. Applied to two driving WMs, the lower rungs reveal a reversal: the model that ranks higher on visual generation quality (L0) ranks lower on action-following (L1-L2), so visual fidelity does not predict the action-robustness a closed-loop verdict depends on.

Christian Oefinger, Finn Rasmus Schäfer, Korbinian Moller +2

Autonomous Vehicle Systems Lab, Technical University of Munich, Garching b. München, Germany

World Models for Autonomous Driving World Model Evaluation Video World Models

Jul 8, 2026cs.AI

Grounding Spatial Relations in a Compact World Model: Instruction Leakage and a Goal-Free Dynamics Fix

Compact world models that condition on a language goal promise to ground relations such as ``put the red block left of the blue block'' using a sparse set of explicit \emph{reference anchors}. We ask when such references actually ground a relation, and identify a trap: a goal-conditioned predictor reaches a striking

0.90

relation-readout accuracy, yet this is \emph{instruction transcription}, not perception. Withholding the goal collapses it to chance (

0.90\!\to\!0.27

, three seeds) and a counterfactual instruction makes the predicted anchors follow the \emph{false} instruction

94.5\%

of the time (true scene

2.3\%

;

N{=}256

). Tested across three settings and a within-task ablation, our central claim characterizes the confound: \textbf{instruction leakage occurs when the scored quantity is transcribable from the instruction (when the instruction names the answer) and is essentially independent of how predictive the non-instruction inputs are.} Our tabletop and the external BabyAI benchmark leak, whereas a Language-Table forward-dynamics world model whose instruction names \emph{referents} does not, until the instruction is augmented to name the direction; and degrading the action never increases leakage, the opposite of what predictor-competition predicts. The diagnosis prescribes the fix: keep the goal out of the dynamics (it belongs to the planner's cost) and supervise the \emph{read} path, recovering genuine, instruction-independent grounding (

0.88

, identical with and without the goal). The detection protocol and remedy apply to any goal-conditioned world model whose instruction names the scored quantity.

Yufeng Wang, Lu Wei, Haibin Ling

1Stony Brook University · 2Westlake University

World Model Learning Visual Grounding World Model Evaluation Embodied World Models

Jul 7, 2026cs.RO

Imagined Rollouts are Kinematic, Not Dynamic: A Diagnosis of Long-Horizon World-Model Failure

Long-horizon failure in world models is conventionally attributed to compounding error, a generic framing that does not distinguish what kind of error compounds. We propose a kinematic-vs-dynamic reframing: world models tend to imagine kinematically rather than dynamically. We operationalize this as the imagined Kinematic-Consistency Error, a per-step diagnostic that measures how far a rollout departs from a closed-form kinematic null, paired with a perturbation protocol that tests whether iKCE responds when physical conditions cross a regime boundary. We instantiate the diagnostic on a released DreamerV3 checkpoint trained on DMC walker-walk, where imagined iKCE runs roughly two orders of magnitude above that of matched real-physics rollouts. Across a friction sweep that crosses the gait-collapse boundary, the model's iKCE stays statistically flat even as the trained policy's reward collapses through the same range, providing the kinematic-not-dynamic signature. The diagnostic distinguishes kinematic from dynamic imagination at horizons longer than the embodiment's gait period.

Finn Rasmus Schäfer, Korbinian Moller, Yuan Gao +3

Autonomous Vehicle Systems Lab Technical University of Munich, Garching b. München, Germany · Data Analytics and Machine Learning Group Technical University of Munich, Garching b. München, Germany

World Models World Model Evaluation Embodied World Models

Jul 6, 2026cs.RO

Deform360: A Massive Multi-view Visuotactile Dataset for Deformable World Models

Predicting object dynamics (i.e., world modeling) is a fundamental challenge for robotic manipulation, and modeling deformable objects presents a particularly difficult case due to their high-dimensional state spaces and complex material properties. While current world models approach this through two distinct paradigms: learning the dynamics over the 2D pixel space or more explicit 3D geometric space. A systematic understanding of their relative strengths and limitations remains elusive due to the lack of diverse, large-scale real-world data. To address this, we present Deform360, a large-scale visuotactile dataset featuring 198 daily-life objects, 1,980 interaction sequences, and over 215 hours of observations from 41 surround-view cameras and bimanual tactile grippers to capture both global motion and contact-induced local deformations. Leveraging a novel markerless visuotactile 3D tracking pipeline to extract dense geometry and motion, we systematically evaluate current state-of-the-art world models, comparing 2D video models against 3D particle models. Finally, we provide a preliminary demonstration indicating the real-world applicability of our dataset by performing robot planning tasks on deformable objects. Our analysis reveals key insights into the trade-offs between structural priors and scalability, providing a solid benchmark for future research in generalizable deformable object-centric world modeling. Project website: https://deform360.lhy.xyz

Hongyu Li, Wanjia Fu, Xiaoyan Cong +11

Brown University, Providence, RI · Columbia University, New York, NY · Massachusetts Institute of Technology, Cambridge, MA

World Models Deformable Object Manipulation Object-Centric World Models World Model Evaluation

Jul 6, 2026cs.CV

Multiplayer Interactive World Models with Representation Autoencoders

We introduce the first multiplayer world model for highly dynamic environments governed by complex physical interactions. Whereas single-player world models treat the other agents as part of the environment, ours conditions on the action streams of multiple agents, learning to attribute changes in the scene to the correct player and to stay coherent under arbitrary combinations of their actions. We study this problem in the game of Rocket League, where players compete and cooperate under fast, tightly coupled dynamics. Trained on 10,000 hours of gameplay collected with publicly available bots, our 5-billion-parameter latent diffusion model generates four-player matches in real time, producing 20 frames per second on a single Nvidia B200 GPU. Although trained only on short clips, its rollouts stay stable far beyond the training horizon: distributional quality holds steady out to five minutes, the longest horizon we measure, and in practice we observe rollouts continuing for hours with no sign of collapse. We systematically investigate the central design choices: the video codec, the generative objective, and the multiplayer conditioning scheme. In addition, we characterize how behavior changes with model and data scale, including the capabilities that emerge and the failure modes that persist. We further develop targeted evaluations that probe the model's physical understanding rather than visual appearance alone. To support continued research on multiplayer world models, we release our dataset, our full training and inference codebase, and a live demo.

Anthony Hu, Václav Volhejn, Adrien Ramanana Rahary +24

1General Intuition · 2Kyutai · ‡École nationale des ponts et chaussées +1

Action-Conditioned World Models Latent World Models World Model Evaluation Interactive World Models

Jul 6, 2026cs.AI

MoP-JEPA: Hard-Assigned Predictor Mixtures for Stochastic JEPA World Models

JEPA world models commonly predict the next latent state with one regressor. Under stochastic transitions, squared and cosine regression return the conditional mean and its normalized direction, respectively: a single compromise that may match no valid successor. MoP-JEPA instead uses

K

hard-assigned heads and a context-only router to produce a finite candidate set in one pass. On held-out OGBench transitions, graph search with single-output predictors succeeds on

0.02

--

0.09

of queries, whereas MoP-JEPA reaches

0.85

. To distinguish useful successors from indiscriminate coverage, we also measure verified-route success (\emph{realroute}), which checks after graph construction whether the proposal contains a path of real transitions. MoP-JEPA leads this same-protocol metric on all three mazes; an MDN attains high raw coverage but predicts many nonexistent edges.

Zhi Song, Ximing Xing, Zhenchao Tang +8

City University of Hong Kong, China · Tencent, China

World Model Learning Latent Dynamics Modeling World Model-Based Planning Joint-Embedding Predictive Architecture Mixture-of-Experts Models World Model Evaluation

Jul 5, 2026cs.LG

Operator-on-F complements value-equivalence: a planning-time diagnostic for latent world models

World-model evaluation for model-based reinforcement learning typically asks whether the learned model predicts reward and value well, which can leave planning-relevant errors in the model's latent rollouts unmeasured. We introduce a complementary diagnostic, operator-on-F, that compares a model's k-step latent pushforward to the environment's on an observable subset F, using the model's own predictor. On a TD-MPC2 size sweep over cheetah-run, reward-prediction error stays within [0.028, 0.091] for every model size - only about 3x variation - so an unnormalized reward-fit check has narrow resolution to distinguish them; the (unnormalized) Bellman residual and reward error themselves have weak relationships with return (Spearman -0.10 and -0.30). Operator error spans 0.28 to 2.62 over the same sizes. At 317M the operator error is 2.62 - an order of magnitude above the 0.28-0.36 cluster - and the planning return collapses to 0.9, while reward-prediction error (0.091) is the highest of the five but stays within the same small [0.028, 0.091] range as the rest of the sweep. The rank correlation between operator error and return loss is -0.90 (anchor-bootstrap 95% CI [-0.90, -0.70] at n=5 sizes; leave-one-out removal of any single size leaves it at -0.80 or stronger). The operator also returns informative, architecture-discriminating estimates in a cross-architecture comparison between TD-MPC2 and a pure-SSL latent world model. The operator diagnostic complements value-equivalence rather than replacing it.

Donna Vakalis

1Mila – Quebec AI Institute · 2Université de Montréal

World Model-Based Planning Latent World Models World Model Evaluation Model-Based RL

Jul 2, 2026cs.RO

GigaWorld-1: A Roadmap to Build World Models for Robot Policy Evaluation

Evaluating embodied robot foundation models remains a critical bottleneck; unlike large language models efficiently assessed via digital benchmarks, robotic policies require slow, costly real-world rollouts limited by hardware and human supervision, which has driven interest in world models as surrogate policy evaluators, yet the key properties that make a world model reliable for policy assessment remain poorly understood. This work presents a systematic study of world models for robotic policy evaluation and introduces WMBench, a benchmark constructed from real-robot teleoperation data and matched policy rollouts covering diverse manipulation tasks to enable controlled comparisons across model families, action encodings, rollout horizons, and evaluation metrics. Using WMBench, we analyze 7 video world models, 4 action representation schemes, and over 324,000 simulated policy rollouts paired with real robot executions, further enriching our analysis with large-scale community submissions from the CVPR 2026 GigaBrain Challenge, curated synthetic trajectories, and a training videos spanning more than 12,000 hours. Our experiments deliver three core insights: evaluator quality is dominated by long-horizon, action-faithful rollout consistency rather than short-term visual realism; pretraining gains stem not only from data scale but from balancing general world knowledge with robot-specific controllability; and architectural choices including action encoding, memory design, and evaluator-focused post-training strongly determine alignment with real-world robot behavior. Drawing on these results, we derive a practical design roadmap and realize it in \textit{GigaWorld-1}, a world model specially optimized for policy evaluation, and we fully release our code, models, datasets, and toolkits to advance scalable evaluation research for embodied foundation models.

GigaWorld Team, Angyuan Ma, Boyuan Wang +24

GigaAI · Tsinghua University

Robot Foundation Models Action-Conditioned World Models Robotic Policy Evaluation Robot Manipulation Benchmarks World Model Evaluation Video World Models

Jul 2, 2026cs.LG

Reward Observability and the Limits of Offline Checkpoint Selection in RSSM World Models

We study the closed-loop properties of a recurrent state-space model (RSSM) world model trained on human demonstrations in Gymnasium's LunarLander-v3. We use the trained world model for zero-shot CEM model-predictive control (MPC) and for actor-critic (A2C) training in imagination. Scored on 100 held-out episodes, the selected model-based A2C policy (trained on world-model checkpoint 280) reaches a mean return of +189.5, matching the best model-free A2C checkpoint (+183.7; 600- and 1000-step episode caps respectively) with ~65x fewer real training transitions. We also compare world-model MPC with a behaviour-cloning (BC) policy trained on the successful demonstrations. The BC policy matches MPC's mean return only under stochastic action selection, and on the same 20 episodes it has one catastrophic episode where MPC has none. We then introduce the Reward Observability Fraction (ROF), the Euclidean fraction of the reward gradient in the observable subspace of the linearized latent dynamics, and show that the next H observations carry Fisher information about every direction in this subspace and none about directions orthogonal to it. ROF itself however depends on how the latent is scaled: rescaling it changes ROF but not the model, so raw levels are not comparable across models. Forcing the reward head onto the posterior-corrected latent z raises ROF, and the rise survives a coordinate-invariant check on the pair of runs we tested. Finally, we test whether ROF or other offline metrics can predict the closed-loop collapse of MPC. Collapse varies between training runs with identical data and configuration. ROF does not predict it, none of the 108 offline summaries we screened passes a permutation test, and the best candidate fails on new runs. Predicting collapse offline from the model and logged data alone remains open.

Nikolai Smolyanskiy, Jonathan Shock

Reinforcement Learning Latent World Models World Model Evaluation Model-Based RL

Jul 1, 2026cs.RO

RoboWorld: Fast and Reliable Neural Simulators for Generalist Robot Policy Evaluation

Video world models are emerging as a scalable alternative for evaluating generalist robot policies, bypassing the physical constraints and engineering burdens of real-world deployment. However, evaluating policies with video world models remains challenging, as world-model errors can make generated rollouts unreliable and slow inference limits large-scale throughput. We introduce RoboWorld, an automated evaluation pipeline that pairs a fast autoregressive video world model with a task-progress-aware vision-language model scoring. To enable reliable long-horizon autoregressive world-model rollouts, we propose Step Forcing, which combines anchored and one-step self-forwarded contexts to reduce train-test mismatch while preserving action-observation dynamics. Together, these components enable RoboWorld to align strongly with real-world robot evaluation across tasks and environments, achieving Pearson's r = 0.989 and Spearman's

ρ

= 0.970.

Byeongguk Jeon, Seonghyeon Ye, JaeHyeok Doo +4

1KAIST · 2Config

World Model Learning Robotic Policy Evaluation World Model Evaluation Video World Models

Jun 30, 2026cs.CV

WorldRoamBench: An Open-World Benchmark for Long-Horizon Stability of Interactive World Models

Despite rapid progress in interactive world models (IWMs), existing benchmarks evaluate action following only at trajectory level and ignore memory and interaction physics. We introduce WorldRoamBench, an open-world benchmark for long-horizon stability across four dimensions, each with tailored innovations: (i) Action: per-frame action metric bypassing cross-model semantic scale disparity and exposing failures hidden by trajectory; (ii) Vision: segment-based drift metric capturing non-monotonic mid-sequence collapse missed by start-vs-end comparisons; (iii) Physics: controllability-gated evaluation over mechanics, optics, and 3D consistency, scoring plausibility under faithful action execution; (iv) Memory: action-decoupled protocol evaluating scene memory via transition-localized 3D point-cloud reconstruction and subject memory via tracking-plus-VLM reasoning. The benchmark comprises 600+ test cases across Nature, Urban, and Indoor scenes in first/third-person views with WASD 10-60s continuous interaction. Evaluating 10+ open/closed-source models reveals none reliably satisfies all dimensions; even the best achieves only moderate scores. Advances on WorldRoamBench are steps toward IWMs that are stable, physically grounded, memory-faithful, and deployable in real-world applications.

Ting-Bing Xu, Jiacheng Sui, Zhe Gao +12

AMAP CV Lab, Alibaba Group · Nanjing University · Tsinghua University +1

World Model Evaluation Interactive World Models

Jun 30, 2026cs.AI

World-Model Collapse as a Phase Transition

Water looks unchanged as it warms, then at a critical point it boils. We ask whether long-horizon language agents show an analogous transition in their implicit world models. In some parameter settings, changing state load by a small amount, or adding a single step of horizon, leaves behavior nearly unchanged; near a critical boundary, the same small change causes a sudden world collapse. We study this effect in a deterministic task family with exact per-step gold state. A large grid search over state cardinality, dependency density, horizon, branching, observation mode, and mutation rate reveals a phase diagram: a solved plateau, a narrow transition band, and a collapse floor. Per-step traces show the mechanism: world-state fidelity fails before action validity, so the agent is not merely choosing a bad action; it is acting from a corrupted world. Stronger models translate the critical boundary but do not remove the qualitative transition. These results make world-model collapse a measurable bottleneck for long-horizon agents.

Xinyuan Song, Zekun Cai

Emory University, Atlanta, GA, USA · The University of Tokyo, Tokyo, Japan · LocationMind, Tokyo, Japan

Phase Transitions Model Collapse Long-Horizon Agent Evaluation Long-Horizon LLM Agents World Model Evaluation

Jun 27, 2026cs.CV

A Physics-Grounded Benchmark for Multi-Agent Dynamics in World Models

Generative world models hold immense promise as scalable simulators for autonomous systems, particularly for synthesizing rare but safety-critical multi-agent interactions, such as vehicle collisions. However, current evaluation paradigms index heavily on visual fidelity and semantic alignment, leaving a critical blind spot: they cannot reliably quantify whether generated dynamics actually obey the fundamental physical laws required for reliable simulation. Assessing this physical plausibility is inherently difficult due to a lack of physical metrics and the challenge of extracting metric-scale kinematics from uncalibrated video rollouts. To bridge this gap, we introduce CrashTwin, a physics-grounded evaluation framework designed to stress-test the physical trustworthiness of world models. CrashTwin couples a diverse dataset of multi-agent collision scenarios, comprising 25K controllable synthetic and 12K in-the-wild real-world collision sequences with a novel calibration-free reconstruction pipeline, enabling the recovery of 3D physical attributes directly from world model rollouts. We propose a diagnostic suite that systematically evaluates three dimensions: spatio-temporal consistency, momentum and kinetic energy conservation, and world-dynamics integrity. Extensive benchmarking of state-of-the-art models reveals a crucial insight: high perceptual quality frequently masks severe physical violations during complex interactions. By quantitatively exposing these failure modes, CrashTwin provides a vital diagnostic tool for developing physically grounded world models capable of reliable real-world simulation.

Nuo Chen, Lulin Liu, Zihao Li +12

Texas A&M University · University of Minnesota · Marquette University +6

Physical Consistency in Video Generation Autonomous Driving Safety Evaluation Multi-Agent Systems World Model Evaluation

Jun 26, 2026cs.RO

Event-Conditioned Diagnostics of Kinematic, Contact, and Object-Permanence Fields in Passive Object-State World Models

World models can predict future physical states, but prediction accuracy alone does not explain how physical information is organized and used inside their latent dynamics. We introduce a controlled diagnostic protocol for studying event-conditioned latent physical structure in passive object-state world models. The protocol tests whether hidden representations encode event-regime information, whether event contexts reweight non-exclusive physical field readouts, and whether field-aligned representational components have functional consequences for prediction. Using a balanced controlled-generator dataset with free-motion, collision, and occlusion events, we evaluate recurrent, attention-based, and latent state-space transition models under a fixed-horizon forecasting setup. The models learn useful predictive dynamics and their hidden states support reliable event-regime readout. Event contexts systematically reweight kinematic, contact, and object-permanence field readouts: free motion is kinematic-dominant, collision combines kinematic and contact structure, and occlusion combines motion-related and object-permanence structure. Time-aligned and directional-consistency analyses further show phase-related shifts in field emphasis. Finally, fixed-horizon projection causal field effect (CFE) shows that suppressing field-aligned directions can degrade event-relevant prediction, with strongest evidence for contact-aligned structure in collision-contact windows and more qualified evidence for object-permanence-aligned structure in hard-occlusion hidden windows. These results support event-conditioned organization and fixed-horizon functional sensitivity of latent physical fields, while not implying explicit physical modules, isolated causal circuits, or context-invariant sliding-window generalization.

Yang Liu, Yuming Chen

College of Intelligent Robitcs and Advanced Manufacturing, Fudan University · Shanghai, China

Latent Dynamics Modeling Latent World Models Object Permanence World Model Evaluation Hidden-State Probing

Jun 26, 2026cs.AI

Understanding Rollout Error in Graph World Models

World models are increasingly used for planning, yet most analyses of rollout error assume vector-valued states and scalar error amplification. Many planning environments, however, are naturally graph-structured: agents, tools, skills, routes, and dependencies interact through evolving relations. In this work, we study how prediction errors accumulate in Graph World Models (GWMs). We formulate fixed-edge and dynamic-edge GWM rollouts under a unified state-action transition framework and derive topology-aware error bounds. For fixed-edge rollouts, we show that long-horizon node error separates into a topology factor, governed by the graph spectral radius, and a model factor, governed by layer spectral norms. For dynamic-edge rollouts, we introduce a joint node-edge error operator that captures feedback between feature prediction and structure prediction, revealing when edge errors amplify future message passing. Motivated by these bounds, we propose Error-Aware GWM, a training objective that combines spectral regularization, rollout consistency, and critical-node weighting. Across synthetic graph topologies and heterogeneous agent-graph testbeds, we find that rollout error and planning regret grow with horizon, that dynamic-edge training is necessary when structure evolves, and that Error-Aware GWM improves long-horizon stability without sacrificing one-step accuracy. Our results characterize when graph world models remain reliable under autoregressive planning and when topology makes them fail.

Xinyuan Song, Zekun Cai

Emory University, Atlanta, GA, USA · The University of Tokyo, Tokyo, Japan · LocationMind, Tokyo, Japan

World Model Learning World Models World Model-Based Planning Dynamic Graph Learning Graph-Based Planning World Model Evaluation

Jun 25, 2026cs.CV

MemoBench: Benchmarking World Modeling in Dynamically Changing Environments

Video generation models aspire to simulate dynamic environments, and several benchmarks now evaluate memory consistency across frames. However, most assess consistency only while the target remains in view, and the few that force objects out of view evaluate static scenes where nothing changes during occlusion. To bridge this gap, we introduce MemoBench, a diagnostic benchmark built around the disappear-and-reappear paradigm in dynamically changing environments: a target object undergoes a physical process, disappears from view, and must be correctly recovered in its updated state upon reappearance. We curate 360 ground-truth clips spanning synthetic and real-world scenes, and design an evaluation suite combining automated metrics with VQA-based assessment across four diagnostic pillars. Evaluation of eight state-of-the-art models reveals key insights and open challenges regarding memory consistency under the disappear-and-reappear paradigm.

Haoyu Chen, Kaichen Zhou, Hang Hua +11

Harvard University · MIT · MIT-IBM Watson AI Lab +5

Physical Consistency in Video Generation World Model Evaluation Video Generation Evaluation

Jun 25, 2026cs.LG

Hallucination in World Models is Predictable and Preventable

Modern generative world models render increasingly realistic action-controllable futures, yet they frequently hallucinate: rollouts remain visually fluent while drifting from the ground-truth dynamics. We hypothesize that hallucination concentrates in low-coverage regions of the state-action space, where lightweight data-centric signals can both detect it and guide mitigation. To test this, we introduce MMBench2, a 427-hour, 210-task dataset for visual world modeling with ground-truth actions, rewards, and live simulators, and train a 350M-parameter world model on it. We identify three distinct hallucination modes: perceptual, action-marginalized, and scene-diverging -- each anchored to a different stage of the pipeline, and develop three signals that accurately predict where the model will fail. To close coverage gaps at training time, we develop a coverage-aware sampling technique; to close them online, our hallucination predictors serve as curiosity rewards for targeted data collection, yielding a data-efficient finetuning recipe that adapts the pretrained world model to entirely unseen environments with as few as 50 real environment trajectories. Overall, our findings reveal that hallucination in world models is inherently a data coverage issue, and that the same signals used to detect it can also be used for mitigation. An interactive web version of our paper is available at https://www.nicklashansen.com/mmbench2

Nicklas Hansen, Xiaolong Wang

1UC San Diego

World Model Learning World Models Action-Conditioned World Models World Model Evaluation

Jun 25, 2026cs.SE

Towards Evaluation of Implicit Software World Models in Coding LLMs

Software engineering, whether performed by humans or by AI agents, requires reasoning about how software behaves. We call the internal model that supports such reasoning the software world model, and view current code-execution benchmarks as covering one well-studied slice of it -- control flow. In this paper, we take a step toward a broader evaluation by shifting the observable axis to execution resources: alongside test outcome and exception class, we predict peak memory, wall-clock time, and ranked profiler outputs at method and line granularity. We use SWE-bench Verified as the source of data to hold the test close to real-world software engineering tasks. All tested models, frontier ones included, show modest performance and brittle behaviour, suggesting a notable lack of understanding of how software is executed, as opposed to how its source code is written.

Egor Bogomolov, Yaroslav Zharov

1JetBrains Research

LLM Evaluation World Model Evaluation

Jun 23, 2026cs.AI

World Models in Pieces: Structural Certification for General Agents

In the big-world regime, agents cannot be universally capable and their ability is inevitably specialized across a world model in pieces. Consequently, standard uniform guarantees fail to distinguish between the understanding of critical bottlenecks and irrelevant failures. We first formalize this limitation by proving that general agents are not universal, rendering standard worst-case analysis uninformative. To overcome this, we introduce structural certification, a transition-local framework that maps bounded goal-conditioned performance to entry-wise guarantees on the agent's internal world model. Our main contribution is constructive. We provide algorithms that filter specific transitions using deep compositional goals and prove that a general agent on these goals has a structural world model with a

\mathcal{O}(1/n) + \mathcal{O}(δ)

error bound. Conversely, this bound is tight in the small-

δ

regime, whose existence is explicitly guaranteed by our certification. These results enable the certifiable deployment of general agents by localizing the specific transitions where long-horizon planning is reliable.

Yikai Lu, Yifei Wu, Xinyu Lu +1

School of Data Science, The Chinese University of Hong Kong, Shenzhen, Shenzhen, China.

Long-Horizon Agent Evaluation Model-Based Planning World Model Evaluation

Jun 23, 2026cs.CL

Qwen-AgentWorld: Language World Models for General Agents

A world model predicts environment dynamics based on current observations and actions, serving as a core cognitive mechanism for reasoning and planning. In this work, we investigate how world modeling based on language models can further push the boundaries of general agents. (i) We first focus on building foundation models for agentic environment simulation. We introduce Qwen-AgentWorld-35B-A3B and Qwen-AgentWorld-397B-A17B, the first language world models capable of simulating agentic environments covering 7 domains via long chain-of-thought reasoning. Leveraging more than 10M environment interaction trajectories of 7 domains in real-world environments, we develop Qwen-AgentWorld through a three-stage training pipeline: CPT injects general-purpose world modeling capabilities from the state transition dynamics and augmented professional corpora, SFT activates next-state-prediction reasoning, and RL sharpens simulation fidelity through a tailored framework with hybrid rubric-and-rule rewards. To evaluate language world models, we present AgentWorldBench, a comprehensive benchmark constructed from real-world interactions of 5 frontier models on 9 established benchmarks. Empirical results demonstrate that Qwen-AgentWorld significantly outperforms existing frontier models. (ii) Beyond foundation models, we further investigate two complementary paradigms through which world modeling enhances general agents. First, as a decoupled environment simulator, Qwen-AgentWorld supports scalable and controllable simulation of thousands of real-world environments for agentic RL, yielding gains that surpass real-environment training alone. Second, as a unified agent foundation model, world-model training acts as a highly effective warm-up that improves downstream performance across 7 agentic benchmarks. Code: https://github.com/QwenLM/Qwen-AgentWorld

Yuxin Zuo, Zikai Xiao, Li Sheng +30

Qwen Team

LLM World Models Agentic RL World Model Learning World Model Evaluation

Jun 23, 2026cs.CV

Trimming the Long-Tail of Visual World Modeling Evaluation

Physical interactions follow a long-tailed distribution: a set of common and regular interactions dominates human experience and visual data, while a broad spectrum of rare and irregular interactions remains underrepresented. Although recent visual world models, including image and video generation models, achieve impressive realism on existing benchmarks, they primarily focus on simulating common physical interactions. This raises a central question: Do current visual world models internalize and generalize physical principles? In this work, we introduce Tailor-Bench, a benchmark that challenges world models to simulate irregular physical interactions. To enable systematic evaluation, we design three scenario modes that progressively challenge model reasoning: Regular scenarios reflect common tool-task pairs, Unconventional scenarios replace conventional tools with attribute-compatible substitutes to test affordance generalization, and Impossible scenarios introduce attribute-violating tools to probe constraint awareness. Additionally, we design two complementary settings under a unified evaluation protocol: predictive generation requires inferring outcomes without guidance, while descriptive generation specifies the target outcome for faithful realization. Our experimental results reveal a clear long-tail gap in physical world modeling: performance degrades from Regular to Unconventional and Impossible scenarios, indicating limited generalization beyond common interactions. Failure analysis further shows that models rely on superficial visual patterns: image models fail to realize correct state changes, while video models further suffer from temporal inconsistencies.

Bingxuan Li, Yining Hong, Cheng Qian +6

University of Illinois Urbana-Champaign · Stanford University · Columbia University

Neural Network Generalization Physical Consistency in Video Generation World Model Evaluation Video World Models

Jun 23, 2026cs.LG

Conformal Orbit-Valid Trust Horizons for Equivariant World Models

Learned world models are useful only over horizons on which their rollout error remains controlled. We study trust-horizon certification for latent world models with known group symmetries. Given a one-step latent residual and a finite-time expansion estimate, we form a raw horizon curve and calibrate it with a split-conformal multiplicative factor. On the reproducible audit set, the conformal factor is

γ_α=1.0

: the raw certificate is already conservative under the audit protocol. Across 50 stable audits, we observe zero anti-conservative violations, corresponding to an exact-binomial 95% upper bound of 5.8% on the violation rate. Our main structural result is that exact equivariance transports a calibrated trust-horizon curve over the group orbit: when the environment dynamics, encoder, predictor, action transform, and latent metric satisfy the stated equivariance/invariance conditions, rollout errors and trust horizons are orbit-constant. Empirically, the implemented models exhibit small orbit-transport residuals, with median 1.1% and maximum 4.1% over 14 orbit audits. The certificate is also non-vacuous (median certified-to-measured horizon ratio 0.67). A certificate-level calibration-cost study shows two complementary regimes. On a symmetric 2D substrate, equivariant, plain, and augmented models are all orbit-valid from a single calibration sector -- no separation, because the substrate already makes non-equivariant baselines approximately orbit-robust. A 3D yaw audit shows the other regime: the equivariant model obtains a one-sector safe and non-vacuous orbit-valid certificate, while healthy non-equivariant baselines pay violation, slack, sharpness, or additional-sector cost. The certificate is a conservative, distributional audit rather than a global reachability guarantee, and certificate-guided subgoal spacing is not confirmed in the current 3D CEM-MPC behavior layer.

Hongbo Wang

Department of Mathematics, Stony Brook University, Stony Brook, NY 11794, USA

Equivariant Neural Networks Conformal Prediction Latent World Models World Model Evaluation

Jun 22, 2026cs.RO

RoboGaze: Evaluating Robot World Models via Structured Vision-Language Analysis

Recent advances in robot world models enable synthetic video generation for embodied prediction and planning. However, evaluating these videos is challenging: visually realistic outputs often violate physical laws, temporal consistency, or task logic, while conventional metrics and monolithic Vision-Language Model (VLM) judges fail to generalize or provide precise diagnostic value. We present RoboGaze, a training-free, multi-agent VLM framework that provides structured, interpretable evaluation for generated robot-manipulation videos. Given a task instruction and video, RoboGaze operates via a three-stage pipeline: task-scene grounding, dimension-specific specialist routing, and critic-based verification. It outputs temporally localized glitch reports categorized under a novel 6-dimension, 30-type robotics-specific taxonomy. To benchmark RoboGaze, we introduce a human-validated dataset of 382 clips spanning simulated and real-world multi-view manipulation. Evaluating eight open-source and proprietary VLM backbones, RoboGaze dramatically outperforms zero-shot baselines, improving description-F1 by up to +43 points and temporal alignment (F1 x IoU) by up to +37 points, closing approximately 85% of the gap to the human ceiling. Furthermore, its critic verifier mitigates the "cry-wolf" false-positive flaw of standard VLMs, lifting clean-clip accuracy from under 25% to over 80%. RoboGaze offers a scalable, highly interpretable diagnostic tool for the rigorous evaluation of robot world models.

Minh-Loi Nguyen, Nghiem Tuong Diep, Hung Khang Nguyen +10

Center for AI Research, VinUniversity · 2VinRobotics · 3DFKI +3

VLM Evaluation Robot Manipulation Benchmarks World Model Evaluation Temporal Consistency in Video Generation