Clone Deterministic 3D Worlds with Geometrically-Regularized World Models

arXiv 25.10 2025 59.8 method

TLDR

Proposes GRWM, a geometrically-regularized world model for high-fidelity cloning of deterministic 3D worlds using temporal contrastive learning.

Reasoning

Strengths: Addresses a specific gap in world modeling (deterministic cloning), provides diagnostic evidence for latent geometry as bottleneck, and introduces a lightweight regularization module. Weaknesses: Limited to deterministic scenarios; abstract cuts off before full method/results; no explicit real-world evaluation.

Read-first score

Read-first score 59.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 43.

Recency 8%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Methodology quality 25%

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

Topical relevance 42%

61.4

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

Reproducibility 25%

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 197.

Keyword Scores

world model

world dynamics prediction

world simulator

interactive world model

generative world model

model-based reinforcement learning world model

video world model

Deep Analysis

Innovations

Identifying that the geometric structure of the latent representation, not the dynamics model, is the primary bottleneck for long-horizon fidelity in deterministic 3D world cloning
Proposing Geometrically-Regularized World Models (GRWM) that apply temporal contrastive learning as geometric regularization to reshape the latent space
Introducing a lightweight geometric regularization module that can be seamlessly integrated into standard autoencoders

Methodology

The authors first conduct diagnostic experiments to quantitatively demonstrate that high-fidelity cloning of deterministic 3D worlds is feasible and that the primary bottleneck is the geometric structure of the latent representation. Building on this insight, they propose GRWM, which applies temporal contrastive learning as a geometric regularization to reshape the latent space of autoencoders, providing a stable foundation for dynamics modeling. The regularization module is lightweight and can be integrated into standard autoencoders.

Key Results

The diagnostic experiment quantitatively shows that the geometric structure of latent representation is the primary bottleneck for long-horizon fidelity. GRWM improves world model fidelity by applying temporal contrastive learning as geometric regularization, offering a simple yet powerful pipeline.

Limitations

The approach is demonstrated only on deterministic 3D worlds (e.g., fixed-map mazes, static space robot navigation), and its applicability to stochastic or open worlds is not addressed.

Resources

Paper