Challenge 02 / Transport / Infrastructure / Real-time AI

Public Safety & Infrastructure

How might we leverage AI and real-time data to improve public safety, optimise traffic management and enhance emergency response, enabling faster, smarter and more coordinated decision-making in urban environments?

During a disruption, conditions can change before an operator has pieced together what happened. The challenge is to turn scattered signals into a timely view of the incident, who it affects and what a team could do next.

Working brief pending EIP confirmation

Scene 02 / Illustrative simulation

The model shows a junction with normal traffic and pedestrian crossings. In the second stage a stalled vehicle blocks a lane and a queue grows upstream. In the third, a proposed diversion branches before the closure and the operator checks capacity and access.

A stalled vehicle blocks a lane at a junction while the queue grows upstream.

Scene key / Sample signals

Approach: 31 km/h in the sample window
Junction: All lanes available
Crossing: Crossing stays clear

An illustrative scenario

When a junction loses a lane

Imagine a stalled vehicle near a busy junction. Speeds drop on one approach and a queue starts to grow. The operator needs to distinguish an incident from an ordinary slowdown, judge what else may be affected, and compare a response before sharing it. The numbers below are invented for the walkthrough.

Monitor traffic

Watch the approaches, junction and crossings as vehicles pass through. Compare speeds and queues across normal signal cycles to distinguish a routine red light from a persistent incident.

Example signal
The sample approach averages 31 km/h, the lanes are available and the queue clears after the signal changes.

Operator choice
Keep the reference interval and check data timestamps before comparing it with a new slowdown.
Detect incident

A stalled vehicle blocks one lane and the queue backs up along the approach. The 3D scene shows the cause. A real operator would use a fresh observation or an independent report to verify a suspected blockage.

Example signal
The sample speed falls to 9 km/h and six vehicles queue upstream. The lane-status report has its own timestamp.

Operator choice
Flag a suspected incident and show which signals support it. Mark an old or conflicting report for review before sharing a confirmed closure.
Coordinate response

A proposed diversion turns away before the blocked lane. Compare the diverted route with the current approach, while the closure remains in effect. Check whether the alternative adds pressure at a crossing or affects access for responders.

Example signal
The simulation shows traffic splitting before the blockage. Show the inputs and assumptions behind any route estimate.

Operator choice
Have the operator decide whether to share an advisory, send the evidence to a response team or wait for more confirmation.

Design considerations

What a prototype could do

These are illustrative ideas for a prototype. EIP has yet to confirm datasets, integrations or judging criteria.

Who uses it

A traffic operator monitors the signals and checks the likely incident. An incident coordinator shares the location, lane status and affected approaches with response teams. Each reader needs to tell a confirmed observation from a model estimate.

Example inputs

Use synthetic speeds, vehicle counts, a small road network and simulated lane reports with timestamps. A sample record might contain a 09:04 speed of 9 km/h, an estimate of six queued vehicles and a lane report timestamped 09:03.

Useful outputs

Show an incident timeline and map, an alert with supporting signals, and a comparison of staying on the original route versus using the diversion. Show input age, uncertainty and the assumptions behind any delay estimate.

What to measure

Measure time to flag a known event, false alarms during ordinary congestion and how often stale data leads to a misleading alert. Compare the queue and delay in the simulation with and without the diversion. Document the baseline and demand assumptions.

Test where the signal can mislead

A red light creates a normal queue
Compare consecutive signal cycles. Check whether the slowdown persists after vehicles normally begin moving again.
One report is stale or conflicts with a sensor
Show the age of each input, mark the closure as unverified and ask for review before suggesting a public advisory.
The diversion is already busy
Check downstream capacity and crossing or responder access. Explain why a diversion may need review even if it avoids the original blockage.

Sample demo sequence

Show the baseline
Start with normal traffic and mark the monitored approach, lanes and crossing.
Replay a stalled vehicle
Show when speeds dropped, where the queue formed and what evidence supports the alert.
Compare two responses
Show the original approach and the potential diversion, then let the operator decide what to share.
Replay one failure case
Delay a report or crowd the alternative route. Show how uncertainty changes the advice.