ARGUS

Inspiration

Cities are built to last for decades, but the people living in them experience the same environment very differently. A route that feels comfortable to one person may be exhausting for someone more sensitive to heat, rain, or crowding. This inspired us to build Argus AI — a human-centric urban simulation platform that helps planners understand how real people may experience a city before major planning decisions are implemented.

What it does

Argus creates a digital simulation of an urban environment and allows planners to stress-test it under different conditions such as temperature, rainfall, population, and crowding.

Instead of modelling every citizen as an identical agent, Argus uses behavioural characteristics such as heat tolerance, rain tolerance, crowd tolerance, walking speed, green-space preference, and transit preference. It then simulates how different citizens move through and react to the city.

The system identifies where problems occur, who is affected, and why, recommends possible interventions such as additional shade or drainage, and allows planners to re-run the simulation to compare the before and after impact.

How we built it

We started by collecting responses from 41 participants across four urban scenarios, producing 164 scenario responses. These responses were converted into six behavioural parameters. We then used a Gaussian-copula approach to generate 1,000 synthetic citizens while preserving relationships observed in our sample, rather than randomly generating independent characteristics.

We combined this population with an OSM-derived city, a Python simulation engine, OpenUSD and NVIDIA Omniverse for 3D visualisation. The simulator models routes, individual environmental exposure, congestion and citizen behaviour, while an advisor identifies problems and proposes supported interventions.

Challenges we faced

Our biggest challenge was turning a small survey into a useful population model without pretending that 41 participants represented an entire city. We therefore treat our current behavioural model as prototype calibration, not a validated predictor of real pedestrian behaviour. Our survey also had confounding between heat, crowding and shade in some scenarios, which taught us how important experimental design is when collecting behavioural data.

Another challenge was connecting multiple layers — real geographic data, synthetic citizens, routing, environmental conditions, interventions, analytics and 3D visualisation — into one coherent simulation loop.

What we learned

The biggest lesson was that a digital twin becomes much more useful when it doesn't just answer “What will this city look like?”, but instead asks:

“How will different people experience it?”

We also learned that AI should not replace the urban planner. Argus works better as a decision-support sandbox: planners change assumptions, stress-test the city, inspect the consequences, apply an intervention and simulate again.

Ultimately, our goal is simple: help planners test the human consequences of urban decisions before those decisions become permanent infrastructure.