4 Spatial Intelligence Uses in Smart Cities, and the Questions They Raise

Cities have long used maps and models. What is new is the pairing of 3D models, live data and AI, so that a city can be simulated, queried and, in some cases, controlled. The phrases used for this are loose: spatial intelligence, world models, digital twins. This article picks four documented examples and reads what each source actually says, including how little it says about results and about the people living in the city.

What do 'spatial intelligence', 'world model' and 'digital twin' mean?

Spatial intelligence. World Labs, the company, says on its about page that it aims to advance 'spatial and physical intelligence', which it describes as 'enabling AI to model the world and reason about objects, places, and interactions across space and time'. It says its world models are 'the foundation of that future' and that its first product builds 'spatially coherent, high-fidelity, persistent 3D worlds'. That is a company's own statement of purpose, not an independent definition.

Digital twin. A virtual model of a physical thing or place that is kept in step with it through data. For factories there is an international framework: the ISO 23247 series covers digital twins of observable manufacturing elements, and NIST researchers describe it as generic enough to adapt to discrete, batch and continuous production. No equivalent single standard defines a city twin, so the term varies from a static 3D map to a live simulation.

World model. In AI usage, a model that represents how an environment behaves so a system can predict or plan. The city examples below mostly use hand-built 3D models and scheduling software, not the generative world models that the term often implies. That difference matters when a headline mixes them.

1. Can AI-timed traffic signals change travel times?

What it is. SURTRAC (Scalable Urban Traffic Control) is a decentralised signal-control system from Carnegie Mellon University. Each intersection works out a schedule for approaching traffic and decides when to change phases, then shares projected outflows with downstream neighbours, repeating roughly once per second.

What was deployed, where, when. The 2013 paper, 'Smart Urban Signal Networks: Initial Application of the SURTRAC Adaptive Traffic Signal Control System' by Stephen Smith, Gregory Barlow, Xiao-Feng Xie and Zachary Rubinstein, describes a pilot at nine intersections in a road network in Pittsburgh, Pennsylvania. It also covers how the system handles unreliable sensor data.

Reported results. The paper's abstract page reports that SURTRAC achieved 'major reductions' in travel times and vehicle emissions compared with the pre-existing signal timings. The page gives no figures and does not say who measured them. We saw specific percentages in search summaries but did not find them on a page we could read, so we do not repeat them.

Open questions. The pilot is small and now more than a decade old. Adaptive signals rely on sensors at junctions, so a city should ask what those sensors capture, whether they detect individual vehicles or people, and how long any data is kept. The abstract page does not address this.

2. What does a city-scale 3D model like Seoul's S-Map do?

What it is. S-Map is the Seoul Metropolitan Government's 3D digital map of the city. Korea JoongAng Daily, reporting on 1 April 2021, says the city built it from about 25,000 aerial photographs and drone footage combined with AI programs, with layers for topography, underground electricity, gas and waterworks pipes, subway routes, real-time road traffic and building exteriors. The city calls it an 'exact replica of the city in digital form'.

Cost and stated uses. The report says the city spent 400 million won (about $354,000), with companies including Naver Labs adding 1.4 billion won. It lists planned uses: simulating how new buildings could alter wind patterns and worsen fine dust, predicting the impact of disasters such as wildfires, and, by 2022, self-driving car simulations and earthquake-impact prediction. It says mobile access to a historical map was planned for September 2021.

Reported results. The report describes launch and intentions, not measured outcomes, and it is a news article from 2021. We did not find an official update confirming which planned features went live.

Open questions. The article notes that 'not all the information is available publicly' and that simulations are run 'for public administration purposes'. It describes no privacy or security measures. Maps that include utilities and real-time traffic raise questions about who can see which layers and how they are protected.

3. What can a district 3D model add, as in Helsinki's Kalasatama?

What it is. Helsinki has built 3D city models, including a model of the Kalasatama district. An Aalto University article (published 7 April 2021, updated 1 July 2021) says a research group developed an app called Palvelututka that combines existing City of Helsinki open data in a browser and shows Kalasatama as a multi-layered environment. In the researchers' framing the 3D model serves as a platform for presenting information from different sources.

Stated uses and limits. The article says research questions include raising the level of detail in an area and adding building interiors, and that interior datasets are not directly or freely available, so the project must produce them itself, for example by laser scanning. It also names the Helsinki Energy and Climate Atlas, which lets residents check the solar-power potential of their home. The app is available only in Finnish. The researchers say the field still needs work on the technical basis for city modelling and on sharing models as data resources.

Reported results. The Aalto article reports no measured results or figures for the Kalasatama model. We could not open the city's own project report or its open-data page, so we do not state its pilot dates or file formats.

Open questions. Building interiors are more sensitive than building exteriors. A city adding interiors to an open model should say what is shown, who consented and who can download it.

4. What is a national digital twin like Virtual Singapore?

What it is. Wikipedia's article says Virtual Singapore was first launched on 3 December 2014 as part of Singapore's Smart Nation drive and was completed in 2022, and that the National Research Foundation, the Singapore Land Authority and the Government Technology Agency co-lead it. We could not read the official programme pages, so this is a secondary account. Wikipedia itself carries a banner, dated August 2026, saying the article may contain AI-generated text that has not been verified.

Data and stated uses. According to that article, airborne aerial photography and LiDAR captured terrain and surface data, and vehicle-mounted laser scans and a mobile mapping survey of about 5,500 km of roads added street-level detail, involving more than 50 terabytes of data. Stated uses include urban planning, infrastructure simulation, transport optimisation, disaster planning and environmental monitoring. It quotes the government saying companies could use it to test driverless cars without placing them on heavy-traffic roads.

Reported results. The article lists intended uses and no measured outcomes. It describes no criticism, privacy safeguards or security risks, which is a gap, not evidence that none exist.

Open questions. A national model that combines survey data from many agencies concentrates a lot of information in one place. Who may query it, how access is logged and how outsiders can test products against it are governance questions we could not check.

How do the four compare at a glance?

What should a city ask before adopting one of these systems?

These questions are ours, not drawn from the sources. They apply to any vendor or agency proposal.

For how we weigh vendor claims against independent findings, see vendor claims vs. verified findings and our methodology. For the same habit applied to benchmark numbers, read reading benchmark claims critically. Another field where demonstrations can run ahead of deployment is robotics; see 10 robotics breakthroughs of 2026. For documented problems with AI systems in use, see AI safety incidents and warning signs of 2026.

What are the limitations of this article?

• We found four examples with a readable source, not five. Several candidate sources were blocked or unreadable, and we dropped claims that appeared only in search summaries.
• Three of the four sources date from 2013 to 2021, so they may not reflect the systems today.
• Virtual Singapore rests on a secondary source with its own reliability warning. The official programme pages were not available to us.
• We could not open the City of Helsinki's own project report or open-data listing, so Helsinki details come only from the Aalto article.
• We tested none of these systems and have no data on their accuracy, cost-effectiveness or effect on residents. Corrections are logged on our corrections page.