How it works
Seven steps from a camera to an approved signal plan.
One loop connects the people who time the signals and the people who ride the jeep. Gemini reads the road, an optimizer proposes timings, and a city engineer decides.
The loop
- 1. Capture. Cameras at pilot intersections send video. The demo uses recorded or mock feeds.
- 2. Perceive. Gemini reads the frames and reports each lane: vehicle types, queues, blockages and standing water.
- 3. Decide. An optimizer proposes timings from that data and the situation, like rush hour or a flood.
- 4. Review. A city engineer sees the plan, the reason behind it and its predicted effect.
- 5. Act. Approved plans run in the simulation. Real signals need city authorization and come later.
- 6. Inform. In a later update, the same live picture will feed guidance for commuters.
- 7. Learn. Outcomes are compared with predictions, and the model is retuned.
How a recommendation is chosen
The optimizer follows simple, explainable rules and shows its reasoning to the engineer who reviews it.
| Condition detected | Action recommended |
|---|---|
| High density on an approach | Extend green for that approach, up to a maximum green cap |
| Low density on an approach | Reduce green and reallocate the time to busier approaches |
| Emergency vehicle detected | Priority green corridor along its route, then restore the normal cycle |
Limits on every recommendation
No plan leaves the optimizer without passing these.
- Minimum green time on every approach
- Fixed yellow and all-red intervals
- Enough time for pedestrians to cross
- A maximum wait, so no lane gets starved
What the AI considers
| Factor | Where it comes from | How it is used |
|---|---|---|
| Traffic jams and queues | Camera feeds, Maps Platform traffic data | Lane priority and timing |
| Rush hours and time of day | Historical counts, schedules | Timing plans by period |
| Types of transport | Gemini vision | People-weighted priority |
| Road size and lanes | Map data, site survey | Capacity and queue limits |
| Road surface condition | Site survey, reports | Speed and capacity adjustments |
| Flooding and weather | Weather data, photos, reports | Closure and reroute scenarios |
| Events | City advisories, schedules | Scenario presets |
| Emergency vehicles | Camera feeds, dispatch input | Priority corridors |
What it is built with
| Part | Technology |
|---|---|
| Video understanding and explanations | Gemini |
| Assistant and tool use | Agent Development Kit |
| Map, markers and traffic data | Google Maps Platform |
| Data, sign-in and approval log | Firebase |
| Signal timing calculation | A standard optimization algorithm with safety limits |
Keep reading
See it for yourself.
The app runs on sample data. Open the demo account and try the dashboard, the sandbox and the assistant.