Skip to content

Source paper

Grounding World Simulation Models in a Real-World Metropolis

Junyoung Seo, Hyunwook Choi, Minkyung Kwon and 10 more

Junyoung Seo, Hyunwook Choi, Minkyung Kwon, Jinhyeok Choi, Siyoon Jin, Gayoung Lee, Junho Kim, JoungBin Lee, Geonmo Gu, Dongyoon Han, Sangdoo Yun, Seungryong Kim, Jin-Hwa Kim

arXiv 2603.15583 · 2026107 upvotes on Hugging Face

Research ideas from this paper (3)

Generated by AI. The paper’s authors did not write these.

AI summary

The paper introduces the Seoul World Model (SWM), a city-scale video generative world model that anchors simulated environments to the actual physical layout of a real metropolis. The main contribution is a retrieval-augmented generation framework that leverages cross-temporal pairing, synthetic urban data, and a Virtual Lookahead Sink to synthesize spatially faithful, temporally consistent, and long-horizon videos.

How it works (AI summary)

SWM fine-tunes a pretrained Diffusion Transformer (Cosmos-Predict2.5-2B) for autoregressive video generation. It retrieves nearby panoramic street-view images based on geographic coordinates, using them for geometric referencing via depth-based forward splatting and semantic referencing via cross-attention, while a Virtual Lookahead Sink mechanism prevents error accumulation over long trajectories.