GeoWorld: Geometric World Models

CVPR 26 2026 45.2 method

TLDR

GeoWorld uses hyperbolic geometry in a world model to improve multi-step visual planning, achieving small gains on CrossTask and COIN.

Reasoning

Strengths: novel hyperbolic latent space addressing Euclidean limitations and long-horizon degradation. Weaknesses: modest improvements (2-3% SR) and limited evaluation on two datasets only.

Read-first score

Read-first score 45.2, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 30.

Recency 6%

100

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

Methodology quality 18%

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

Reproducibility 18%

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

Topical relevance 29%

42.9

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%

25.4

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.25410212

Citation velocity 12%

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 366.

Keyword Scores

world model

model-based reinforcement learning world model

world dynamics prediction

world simulator

interactive world model

video world model

generative world model

Deep Analysis

Innovations

Hyperbolic JEPA that maps latent representations from Euclidean space onto hyperbolic manifolds to preserve geometric structure and hierarchical relations
Geometric Reinforcement Learning for energy-based optimization in hyperbolic latent space, enabling stable multi-step planning

Methodology

GeoWorld introduces a Hyperbolic JEPA that transforms latent representations from Euclidean space into hyperbolic manifolds, capturing geometric and hierarchical structure among states. It then employs Geometric Reinforcement Learning for energy-based optimization, allowing stable multi-step planning directly in the hyperbolic latent space. The model is evaluated on CrossTask and COIN datasets against the V-JEPA 2 baseline.

Key Results

GeoWorld achieves approximately 3% success rate improvement in 3-step planning and 2% improvement in 4-step planning compared to the state-of-the-art V-JEPA 2 on CrossTask and COIN.

Resources

Paper Project