Paper summary: ABot-Earth 0.5: Generative 3D Earth Model

ABot-Earth 0.5 generates large, seamless 3D Gaussian Splatting environments from satellite images alone, at under 10 minutes per square kilometre, viewable in web map engines and usable as a simulator for UAV navigation.

Figure from the paper "ABot-Earth 0.5: Generative 3D Earth Model"

Figure from Qian et al. (2026), licensed CC BY 4.0. Figure 1 : We are unveiling ABot-Earth 0.5 , a generative 3D Earth model. Our official launch showcases an evolving 3DGS world spanning over 300 cities across 190+ countries, with continuous global expansion.

Problem

Building realistic 3D models of whole cities usually takes expensive capture and reconstruction. Simulation for embodied AI (drones, robots) needs such environments at scale.

Method

ABot-Earth 0.5 is a generative model formulated directly in 3D Gaussian Splatting (the 2023 3DGS paper is the first one summarised on this blog). It’s trained on many existing real-world urban reconstructions to learn realistic geometry and textures. At inference it generates new 3D scenes conditioned only on satellite imagery. A hierarchical level-of-detail structure allows real-time viewing on web map engines.

Key results

From the abstract and Figure 1:

synthesis at under 10 minutes per square kilometre, with high realism
a launch showcase spanning 300+ cities in 190+ countries
used as a simulation sandbox that narrows the sim-to-real gap, for example in closed-loop UAV navigation

Limitations

There’s no dedicated limitations section in the extracted text. Generated geometry is a plausible reconstruction from overhead imagery, not a survey-grade capture.

Paper summary: ABot-Earth 0.5: Generative 3D Earth Model

Problem

Method

Key results

Limitations

Sources