Real traffic, from Google Maps. Four roads, one set of lights.
Two minutes. Keep all four queues short.
Delhi is making 1,092 junctions adaptive.
TraffiCure measures every corridor before and after, on the Google Maps data that already exists.
Each road is modelled from TraffiCure’s Google Maps Platform feed for this junction. Two-minute travel times from 13–28 September 2026 set how congested each approach gets at each time of day. Hourly vehicle counts from June to August 2026 set how many vehicles arrive, and where they go.
Ring Road’s through traffic runs on the RK Puram flyover and isn’t signalised; you control the at-grade junction underneath. The flyover is drawn between rounds and dissolves while you play, so nothing hides the lights; the traffic on it is scenery and never enters the counts. TraffiCure AI plays the same traffic on its own copy of the junction: before every change of lights it tries each option (every other road, or holding the green a little longer), played out with the traffic already heading its way, and picks the one with the least waiting. Between games it works the next round out ahead of time, starting from the ways a careful visitor would play it and trying a couple of hundred small changes to its plan. Like a real signal controller, it never keeps a road with traffic waiting red for more than 4 minutes; otherwise it plays for the least waiting over the whole round, so its score can dip early while it clears a long queue. Under today’s fixed signal timing, the simulated junction reproduces Google’s observed approach delays by time of day. While you play, the green count on each light is how many vehicles the AI has waiting there on exactly the same traffic (the booth panel can also draw where its queue ends on the road).
Source: TraffiCure production database, Delhi network. Segment IDs from the Google Roads Management Insights feed.