ExtraGS: Geometric-Aware Trajectory Extrapolation with Uncertainty-Guided Generative Priors
Kaiyuan Tan Yingying Shen Haohui Zhu Zhiwei Zhan Shan Zhao, Mingfei Tu Hongcheng Luo Haiyang Sun Bing Wang, Guang Chen Hangjun Ye · 2025
Read ExtraGS: Geometric-Aware Trajectory Extrapolation with Uncertainty-Guided Generative Priors by Kaiyuan Tan Yingying Shen Haohui Zhu Zhiwei Zhan Shan Zhao, Mingfei Tu Hongcheng Luo Haiyang Sun Bing Wang, Guang Chen Hangjun Ye (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-15529.txt.
Description
Kaiyuan Tan and colleagues improved virtual test drives in 2025 by rendering what a car would see on roads it never quite drove. Simulators extrapolate new viewpoints from recorded driving logs, but drifting away from the recorded path exposes artifacts, and generative models used as stand-in ground truth tend to blur geometry. ExtraGS, their framework, insists on both geometric discipline and generative help, each where it belongs.
A hybrid representation marries Gaussian splatting to a signed distance function for the road surface, keeping the drivable plane exact, while far-field Gaussians with learnable scales handle distant scenery cheaply. A self-supervised uncertainty estimate, built from spherical harmonics, flags where extrapolation is actually breaking, and only there are generative priors invited in. Across several datasets and multi-camera rigs, the extrapolated views gained realism and geometric consistency while keeping the fidelity of the original recordings.
Catalog record
| CREATOR | Kaiyuan Tan Yingying Shen Haohui Zhu Zhiwei Zhan Shan Zhao, Mingfei Tu Hongcheng Luo Haiyang Sun Bing Wang, Guang Chen Hangjun Ye |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-15529