Traffic congestion remains one of the most pressing challenges for South African cities. Fixed-time signal plans, designed for average conditions, struggle to cope with the variability of real-world traffic — morning surges, afternoon school runs, unexpected incidents, and event-day demand. Quebit AI was built to solve exactly this problem.
How Quebit Works
Quebit is Syntell’s AI-driven adaptive signal coordination platform. It sits on top of existing controller infrastructure (MX, XTE, or compatible third-party hardware) and continuously reads live traffic data from detectors at each intersection. Using machine learning models trained on historical and real-time patterns, Quebit calculates optimal signal timings and pushes updated plans to controllers in real-time.
The system operates at two levels:
- Intersection-level optimisation — adjusting green splits, phase sequencing, and cycle lengths at individual intersections based on current queue lengths and approach volumes
- Corridor-level coordination — synchronising green waves across a sequence of intersections to create smooth progression for the dominant traffic flow, reducing stop-and-go driving
Pilot Results
The first Quebit pilot was deployed across a 3.2 km corridor of 8 signalised intersections. Over a 12-week measurement period comparing Quebit-managed operations against the previous fixed-time plans, the results were clear:
- 30% reduction in average peak-hour corridor travel time
- 22% fewer stops per vehicle traversing the full corridor
- 18% improvement in intersection level-of-service during the PM peak
- Consistent performance across both typical weekdays and atypical demand events
“What impressed us most was how Quebit adapted to a major incident on a parallel route. Within minutes, it detected the spill-over demand and adjusted our corridor timing — something that would have taken our control room 20 minutes to respond to manually.” — Traffic Engineer, Pilot Municipality
The Technology Behind It
Quebit’s AI engine uses a combination of reinforcement learning and model predictive control. The reinforcement learning component learns optimal strategies over time by exploring different timing configurations and observing their impact on traffic flow. The model predictive control layer ensures that short-term decisions account for predicted demand over the next few signal cycles — not just the current moment.
All processing runs on Syntell’s cloud infrastructure, with results pushed to field controllers via the SIMS platform. This means no additional hardware is required at the intersection — Quebit works with existing detection and communication infrastructure.
What’s Next
Following the success of the pilot, Quebit is being expanded to additional corridors and is available for deployment across any municipality running Syntell-connected controllers. Future development will incorporate pedestrian and cyclist demand into the optimisation model, as well as public transport signal priority.