The year/Independent research

Paper 2606.09967

ABot-Earth 0.5: Generative 3D Earth Model

Published
Jun 2026
Research lab
Independent
Citations
1
GitHub
192 stars

01 In brief

Summary

ABot-Earth 0.5, developed by AMAP CV Lab (Alibaba Group), is a generative 3D framework that synthesizes vast, seamless 3D environments from geospatially referenced satellite imagery using a native 3D Gaussian Splatting (3DGS) representation.

Trained on real-world urban reconstructions, it generates realistic geometry and textures at under 10 minutes per square kilometer, with hierarchical level-of-detail (LOD) for real-time web-based visualization.

The framework aims to mitigate the sim-to-real gap for Embodied AI applications like UAV navigation.

Key innovations include a native 3DGS generative framework, inherent multi-LOD decoding, seamless sliding-window inference, and cross-domain adaptation for robust conditioning.

The data pipeline collects multi-source imagery (satellite, aerial, urban), reconstructs city-scale 3DGS scenes via ABot-3DGS, and curates training tiles.

Deployment uses a tile-based production pipeline (A100 GPUs, 4K resolution per tile, ~312,500 tiles for global built-up area) and the EarthScape rendering pipeline, achieving real-time rendering of ~3.2 trillion Gaussian primitives.

Evaluation shows state-of-the-art FID of 16.1, outperforming baselines (CityDreamer, GaussianCity, EarthCrafter).

Compared to Google Earth, it offers broader coverage (e.g., Ireland) and higher aesthetic scores, though lower geometric/textural fidelity.

A hybrid approach integrates high-fidelity landmark reconstructions (e.g., Eiffel Tower) into generated scenes.

The project aims to democratize 3D content production and enable planetary-scale digital twins, with future work focusing on street-level detail and scaling laws validation.

02 From the paper

Abstract

We present ABot-Earth 0.5, a generative 3D framework designed to synthesize vast, seamless 3D environments from ubiquitous, geospatially referenced satellite imagery. To achieve this, we propose a novel generative model formulated directly with the 3D Gaussian Splatting (3DGS) representation. The model is trained on a diverse corpus of existing real-world urban reconstructions, learning to generate realistic geometry and textures. At inference, it synthesizes novel 3D scenes conditioned solely on satellite imagery at a scalable rate of under 10 minutes per square kilometer, while demonstrating exceptional realism. The framework is designed for accessibility, with integrated hierarchical level-of-detail (LOD) structures that permit real-time, interactive visualization on web-based map engines. This high-fidelity simulation sandbox effectively mitigates the sim-to-real domain gap, enabling critical downstream Embodied AI applications like closed-loop UAV navigation. By providing an ultra-low-cost and high-efficiency solution, ABot-Earth 0.5 significantly lowers the technical and financial barriers to large-scale 3D reconstruction and empowers the future of global digital earth visualization.