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

arXiv 2026 38.6 method, theory

TLDR

World models fail long-horizon due to kinematic (not dynamic) errors; a diagnostic (iKCE) reveals this in DreamerV3 on DMC walker-walk.

Reasoning

Strengths: Novel kinematic-vs-dynamic reframing with a clear diagnostic (iKCE) and empirical validation on a released checkpoint. Weaknesses: Limited to a single environment and checkpoint; no real-world or diverse-domain experiments.

Read-first score

Read-first score 38.6, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 45.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Topical relevance 29%

64.3

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=none

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=checkpoint

Citation impact 18%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 259.

Keyword Scores

world model

model-based reinforcement learning world model

world dynamics prediction

generative world model

world simulator

interactive world model

video world model

Deep Analysis

Innovations

Proposes a kinematic-vs-dynamic reframing of long-horizon world-model failure, distinguishing the type of compounding error.
Introduces the imagined Kinematic-Consistency Error (iKCE), a per-step diagnostic measuring departure from a closed-form kinematic null.
Develops a perturbation protocol that tests whether iKCE responds when physical conditions cross a regime boundary.

Methodology

The method computes imagined rollouts from a released DreamerV3 checkpoint on the DMC walker-walk task, calculates iKCE as the deviation from a closed-form kinematic model, and compares it to real-physics rollouts. A friction sweep is used to cross the gait-collapse boundary, and the response of iKCE versus reward is analyzed.

Key Results

Imagined iKCE is about two orders of magnitude higher than in real-physics rollouts; across a friction sweep, iKCE remains statistically flat even as the policy's reward collapses, indicating the model imagines kinematically rather than dynamically.

Resources

Paper