PanoWorld: A Generative Spatial World Model for Consistent Whole-House Panorama Synthesis

arXiv 2026 46.5 method, application

TLDR

PanoWorld generates consistent whole-house VR panoramas from floorplans using autoregressive node-based generation with 3D shell and Gaussian Splatting cache.

Reasoning

The paper presents a novel method for consistent multi-room panorama synthesis, effectively combining a 3D geometric proxy with a dynamic Gaussian Splatting cache to maintain cross-view coherence. However, it is narrowly focused on VR tour generation and does not demonstrate general world modeling capabilities or include explicit real-world empirical evaluations in the abstract.

Read-first score

Read-first score 46.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 25.

Recency 6%

100

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

Citation impact 18%

82.6

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.82647989

Methodology quality 18%

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

Reproducibility 18%

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

Topical relevance 29%

35.7

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 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: 313.

Keyword Scores

world model

generative world model

world simulator

interactive world model

video world model

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Autoregressive generation of node-based 360-degree panoramas matching discrete VR tour navigation
Floorplan-derived 3D shell as a global geometric proxy for spatial consistency
Dynamic 3D Gaussian Splatting cache as renderable spatial memory
Feed-forward panoramic LRM designed for metric-scale multi-room 360-degree inputs
Room-aware Group Attention to suppress cross-room feature interference
Topology-aware progressive caching strategy for fusing local updates without full history reconstruction

Methodology

PanoWorld treats whole-house panorama synthesis as autoregressive generation of node-based 360-degree panoramas, using a floorplan-derived 3D shell as a global geometric proxy and a dynamic 3D Gaussian Splatting cache as renderable spatial memory. A feed-forward panoramic LRM lifts generated panoramas into local 3DGS updates, while Room-aware Group Attention handles cross-room feature interference, and a topology-aware progressive caching strategy fuses these updates without repeatedly reconstructing the full history.

Key Results

By decoupling shell-based geometry guidance from cache-rendered visual memory, PanoWorld preserves high-frequency 2D synthesis quality while improving cross-node layout and material consistency.

Resources

Paper