From fixed timing to adaptive control
Until the mid-2000s, traffic signal timing in Singapore was managed through pre-set plans — fixed green-phase durations for different times of day, updated quarterly from manual traffic counts. The system worked tolerably when travel patterns were predictable, but struggled with incidents, large events, and the growing complexity of mixed-mode traffic as bus rapid transit corridors were introduced.
The Land Transport Authority (LTA) began rolling out vehicle-presence detectors — inductive loops embedded in road surfaces — from 2008, giving signals the ability to extend green phases when vehicles remained in detection zones. This addressed queue clearance but not network-level flow optimisation; each junction still operated independently.
The critical architectural shift came with the deployment of the Adaptive Traffic Control System (ATCS) from 2014 onward. ATCS introduced a centralised optimisation engine that calculates green-phase timings for groups of adjacent junctions simultaneously, using real-time vehicle count data from overhead cameras rather than in-ground loops. By 2022, ATCS covered more than 1,000 signalised junctions across the island's primary and secondary road network.
The camera network
LTA operates approximately 5,000 traffic cameras at junctions and road segments across the island. The cameras serve dual purposes: they feed vehicle count and speed data into the ATCS optimisation engine, and they provide the visual feed for the TrafficSmart monitoring dashboard used by LTA's Traffic Management Centre.
Camera footage is not stored beyond 24 hours for traffic management purposes; longer retention is restricted to recorded incidents under a separate legislative framework. The image processing pipeline extracts numerical outputs — vehicle count per minute, average speed, queue length in vehicle lengths — and passes these values to the ATCS engine rather than transmitting raw video to central servers, reducing both bandwidth and privacy exposure.
A third function of the camera network, added from 2019, is large-vehicle detection at junctions adjacent to bus stops. When an articulated bus is detected approaching a red phase, the system can grant a two-second green extension to avoid a full signal cycle stop, which is disproportionately time-costly for buses carrying 80–120 passengers.
Predictive modelling and the Traffic Management Centre
Beyond real-time adaptive control, LTA operates a predictive layer that models network conditions 30 to 90 minutes ahead using historical patterns and current inputs. The model ingests live counts from the camera network, real-time ERP (Electronic Road Pricing) transaction volumes, GPS traces from the national taxi and private-hire vehicle fleet (under data-sharing agreements with Grab and Gojek), and bus GPS data from the Land Transport Datamall API.
The output is used primarily for proactive signal plan adjustments ahead of known high-demand periods — stadium events, public holidays, major school examinations — and for routing recommendations pushed to navigation applications via the OneMap API. The model's accuracy at the 60-minute horizon, measured against actual junction throughput, is reported at approximately 83% in LTA's 2023 annual report on intelligent transport systems.
Electronic Road Pricing and demand management
ATCS and camera-based detection manage supply — how efficiently existing road capacity is used. The Electronic Road Pricing system, in operation since 1998 and transitioning to the satellite-based ERP 2.0 from 2024, manages demand by pricing entry into congested corridors at rates that vary by time of day and measured traffic conditions.
The ERP 2.0 transition is particularly relevant to this account because the satellite-based system eliminates the overhead gantry infrastructure of the original ERP, replacing it with onboard units that report vehicle position continuously. This produces a real-time vehicle location dataset — anonymised and aggregated — that feeds directly into the predictive modelling layer, substantially improving coverage in areas between the fixed gantry positions of the original system.
Autonomous vehicle testing infrastructure
A discrete but documented layer of Singapore's traffic technology infrastructure supports autonomous vehicle trials. Since 2017, the LTA and the Singapore Economic Development Board have designated specific road corridors — concentrated in one-north, Punggol Digital District, and Jurong Lake District — as approved AV testing zones. These corridors are equipped with roadside communication units (RSUs) that broadcast real-time signal phase and timing data to test vehicles using C-V2X (Cellular Vehicle-to-Everything) protocol.
The RSU deployment on AV test corridors represents a preview of infrastructure that LTA has stated will roll out to the broader network as AV volumes increase. The current deployment covers approximately 20 kilometres of road in the three designated districts, with 142 RSU units installed as of Q1 2025.
Limitations and documented shortcomings
ATCS optimises for vehicle throughput at the junction level; its objective function does not include pedestrian crossing duration or cyclist wait times. An academic study from the NTU School of Civil and Environmental Engineering (2022) found that ATCS-managed junctions had pedestrian crossing intervals 8–12% shorter during peak vehicle-demand periods than fixed-timing predecessors, an unintended consequence that LTA has acknowledged and is addressing through a pedestrian-flow-inclusive optimisation update expected in 2026.
A second limitation is that ATCS coverage, while broad, does not extend to the full network. Approximately 2,800 signalised junctions in lower-traffic residential areas and industrial estates remain on updated fixed-timing plans rather than adaptive control, on the basis that the throughput benefit does not justify the infrastructure cost at low vehicle volumes.