GeoSem-WAM: Geometry- and Semantic-Aware World Action Models

arXiv 2026 53.6 method

TLDR

GeoSem-WAM enhances world action models by adding geometric and semantic supervision to latent representations, improving action prediction without explicit future rollout.

Reasoning

The paper addresses a key limitation of RGB-only world models by introducing auxiliary prediction branches for geometry and semantics, which strengthens latent representations. Its strengths include a clear motivation and demonstrated improvements in action prediction and robustness. However, the abstract lacks explicit mention of real-world experiments, and the reliance on simulated environments may limit generalizability claims.

Read-first score

Read-first score 53.6, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 37.

Recency 6%

100

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

Citation impact 18%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.9201906

Methodology quality 18%

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

Topical relevance 29%

52.9

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

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Citation velocity 12%

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

Field roles

FoundationFrontierBridge

Rank sensitivity

Stability: volatile; rank range: 400.

Keyword Scores

world model

world dynamics prediction

interactive world model

model-based reinforcement learning world model

video world model

generative world model

world simulator

Deep Analysis

Innovations

Introduces two auxiliary prediction branches for future geometry and semantic representations alongside future RGB prediction
Jointly captures scene dynamics, spatial geometry, and semantic context within a unified latent space
Preserves efficient inference by avoiding explicit future rollout or video generation at test time

Methodology

The proposed GeoSem-WAM framework enhances latent representations through geometric and semantic supervision. It adds two auxiliary prediction branches for future geometry and semantic representations to the standard future RGB prediction, training them jointly to capture scene dynamics, spatial geometry, and semantic context in a unified latent space. The model avoids explicit future rollout or video generation at test time, maintaining efficient inference.

Key Results

Extensive experiments demonstrate that incorporating structured world supervision consistently improves action prediction accuracy, scene understanding, and robustness under challenging embodied scenarios.

Resources

Paper