LaWM: Least Action World Models for Long-Horizon Physical Consistency from Visual Observations

arXiv 2026 61.7 method

TLDR

LaWM uses the Principle of Least Action to govern latent transitions, ensuring long-horizon physical consistency in learned world models from visual observations.

Reasoning

The paper introduces a novel method integrating physical principles directly into latent dynamics, addressing compounding errors in long-horizon rollouts. However, the abstract lacks mention of real-world experiments or benchmarks, and the evaluation scope is unclear.

Read-first score

Read-first score 61.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 47.

Methodology quality 18%

100

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=baseline,benchmark,evaluation,metric,result

Recency 6%

100

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

Topical relevance 29%

67.1

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

Citation impact 18%

66.6

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.66597141

Reproducibility 18%

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

Citation velocity 12%

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

Field roles

FrontierBridgeMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 409.

Keyword Scores

world model

world dynamics prediction

generative world model

world simulator

model-based reinforcement learning world model

interactive world model

video world model

Deep Analysis

Innovations

Operationalizing the Principle of Least Action in learned visual latent space for world modeling
Latent variational integrator that encodes observations into learned generalized coordinates and learns a discrete Lagrangian
Transition rule defined by solving discrete integration condition from variational principle, replacing unconstrained neural transition predictors

Methodology

LaWM encodes visual observations into learned generalized coordinates, learns a discrete Lagrangian over consecutive latent states, constructs a discrete action functional, and advances prediction by solving the corresponding discrete integration condition. This provides a structure-preserving bias for long-horizon visual prediction without relying on auxiliary losses or separate dynamics modules.

Key Results

Across physics-clean synthetic dynamics and embodied robot interaction benchmarks, LaWM improves physical invariance, background consistency, motion smoothness, and appearance and geometric prediction metrics over video-generation and world-model baselines.

Limitations

The method's reliance on learned generalized coordinates may limit applicability to highly complex or unstructured visual scenes.
The discrete variational integrator may introduce approximation errors for very long rollouts or chaotic dynamics.
Evaluation is limited to synthetic and controlled robot interaction environments; real-world generalization is not yet demonstrated.

Resources

Paper