LOCUS: Laboratory of Computational Urbanism and Sustainability
We study how geographic space can be represented, reasoned about and acted on by computational systems, working across GeoAI, spatial tokenization, spatial reasoning and computational urbanism.
Point at the street and people will gather there.
locus plural loci
Locus (plural loci) is a specific place, site or locality. It is also our name: LOCUS Lab is the Laboratory of Computational Urbanism and Sustainability. Both reflect what we study — geographic space, and the computational methods, from GeoAI and spatial reasoning to computational urbanism, that help us understand places and make them more sustainable.
Machine learning models that take geographic context, location and spatial structure as first-class inputs.
Whether and how models infer distance, direction, topology and containment — and where they fail.
Ways of encoding places, geometries and their relationships so that models can learn from them.
Turning continuous geographic space into discrete, learnable units that carry explicit spatial relations.
Cities as data: street networks, parcels and urban form analysed with computational methods.
Applying spatial intelligence to questions about how cities and regions can develop sustainably.
GeoSpatialBench evaluates multimodal models on 8,400 instances from real OpenStreetMap data in six cities — from recognizing spatial relations to editing geometry under explicit constraints.
A geospatial tokenization approach that encodes spatial relationships explicitly in spatial tokens, using road-delimited urban parcels as the basic units.
A guide to GeoAI written chapter by chapter by the lab, alongside shorter notes on methods and ideas.
Open the wikiRecent notes