CASE STUDY · KAUNAS
Less waiting on
every road in.
On Šeštokų g. in Kaunas, cars waited about 12 minutes for a green. We counted every vehicle, rebuilt the junction as a digital twin and let artificial intelligence search for a better plan. Same cars, same morning: the wait fell to under 3 minutes.
CITY PLAN
OUR PLAN
76
side-street cars an hour get through, of ~138
138
side-street cars an hour get through, 82% more
12.2 min
average wait on the side street
2.8 min
average wait on the side street, 77% less
43.8%
of the average trip spent moving freely
60.6%
of the average trip spent moving freely, 39% more
18.0
km/h average speed, all vehicles
27.2
km/h average speed, all vehicles, 51% faster
01 · WHAT CHANGED
Same junction, new order.
Our artificial intelligence rebuilt the order of the lights. It searched every safe combination in the digital twin and settled on four stages instead of five, so Šeštokų g. no longer waits at the back of the cycle.
02 · THE DATA
Computer vision sensors counted it car by car.
Computer vision sensors count every vehicle on every lane, minute by minute and by type. One measured hour as an example: 1,143 vehicles northbound and 596 southbound on the main road.
The bursts every other minute are the lights, visible in the raw counts. The second source is the city's own controller programme: every set of lights, its rules and its stage order.
1,143
northbound vehicles in the example hour
596
southbound vehicles in the example hour
14
sets of lights in the city's programme
5
camera views counted, two at this junction
03 · THE DIGITAL TWIN
A digital copy of the street, for artificial intelligence.
A digital twin is a computer copy of the junction: the same lanes, the same lights, the same cars. Our artificial intelligence tries signal plans on it against real measured traffic, in minutes, without touching the street. The twin has four parts.
ROAD
The junction, lane by lane. We started from OpenStreetMap and corrected it against the real street: crossings and stop lines moved to where they are, right-turn-only lanes marked.
CONTROLLER
The city's traffic light controller, rebuilt as software that follows the same rules: 14 sets of lights, minimum greens, the safety gaps, stages skipped when no one waits, pedestrian buttons. Tested on its own first.
TRAFFIC
A measured hour, replayed vehicle by vehicle, about 2,380 of them, with the neighbouring junction on its real programme too. Every plan faces exactly the same traffic.
SCORING
How a plan is judged. "Flow" is the share of each trip spent moving freely rather than stopped or crawling, averaged over all cars and for the worst-off 10%. Plus speed, cars served and waits on every road. The same plan always scores the same.
04 · THE SEARCH
Millions of possibilities. Only one works best.
A junction like this can be run in a huge number of ways: which flows get green together, in what order, for how long. Nearly all of them leave some road waiting. Tuning the seconds alone was not enough: on the city's own stages an optimiser tried 17 timing changes and every one still starved the side street.
So the search went wider, to the stages themselves. It listed every group of flows that may safely share a green, built every plan from those groups, and tested the sixteen strongest in the twin for a full hour each. One order beat all the others.
49
SAFE GROUPS OF FLOWS, 11 AS LARGE AS POSSIBLE
30
PLANS THAT FIT ONE CYCLE
16
SIMULATED FOR A FULL HOUR
1
BEST ORDER: CROSSINGS AT ONCE, SIDE STREETS TOGETHER
05 · THE RESULT
The numbers, road by road.
Both plans scored on the same simulated hour of the measured morning: the city's plan as it runs today, and our new stage order with its longer cycle.
On Šeštokų g. the average stop falls from about 12 minutes to under 3, and every car that arrives gets through. The main road's busy south arm stops 73% less and moves 108 more cars an hour. No road waits longer.
PLAN
MEAN FLOW
WORST-OFF 10%
SPEED, km/h
SIDE STREET, cars/h
SIDE-STREET WAIT
City plan · today
43.76%
23.16%
18.0
76 of ~138
733 s
Our plan · new order
60.61%
35.65%
27.2
138 of 138
165 s
Change
+39%
+54%
+51%
+82%
−77%
ROAD
CITY PLAN
OUR PLAN
CHANGE
SERVED, CITY → OURS
Main road, north arm · K1/2
39 s
27 s
−30%
543 → 542 cars/h, −0.2%
Main road, south arm · K5/6
134 s
36 s
−73%
1,384 → 1,492 cars/h, +8%
Šeštokų g. · K3/4
733 s
165 s
−77%
76 → 138 cars/h, +82%
North-west side street · K7/8
190 s
85 s
−55%
121 → 121 cars/h, 0%
> NEXT
Have a junction nobody can explain?
We start the same way every time: count it, rebuild it, test the plan before it reaches the street.
TALK TO US →