Description
Ask an LLM which parks are near a building or which road is closest, and it knows almost nothing about real urban space. SpatialLLM from Wuhan University is a framework that enhances LLMs with urban spatial intelligence, turning point clouds, OpenStreetMap and multi-view images into structured scene text.
LLMs can then handle complex urban spatial reasoning; published in JAG 2026.
Multi-source:Point clouds, OSM and images.
Scene text:Structured descriptions.
Reasoning:City-scale Q&A.
LLMs can then handle complex urban spatial reasoning; published in JAG 2026.
Features
Multi-source:Point clouds, OSM and images.
Scene text:Structured descriptions.
Reasoning:City-scale Q&A.
