Timing Sensitivity in Actuated Traffic Signal Control: A simulation study on one urban network

1 Introduction

• The motivating question is whether varying maximum green and fixed-plan duration changes the comparison on the same traffic.The paper cautions that leaving settings unchanged as demand increases can make a timing effect appear to be a controller-family effect.Source: §1 · p. 1
• The study compares eight configurations on one urban network while holding the network, journeys, and other controller parameters constant.The analysis includes four actuated maximum-green settings, three fixed plans, and an actuated variant that ends green when discharge is absent.Source: §1 · p. 1
• Prior timing guidance and phase-termination studies motivate the comparison, but absence of departures alone is not treated as evidence of spillback.The cited precedents include,,, and.Source: §1 · p. 2

2.1 Network, traffic and simulation

• The central Tel Aviv OpenStreetMap network supplies the geometry, while synthetic trips use entry queues and physical removal of vehicles stationary for 400 s.Routes remain fixed and the journey mix is generated from the connected network.Source: §2.1 · p. 2

2.2 Controller configurations

• Fixed plans use 40, 60, or 90 s of total green, while responsive policies share 60 s splits and terminate green by presence or a no-discharge rule.The truncating configuration retains maximum-green multiplier 2; its 4 s no-discharge check applies only after the 7 s minimum green.Source: §2.2 · pp. 2–3, Table 1 · p. 3

2.3 Campaign and pairing

• Ten paired demand seeds across eight configurations and three arrival rates yield 240 runs, measured over a common ten-hour horizon in an exploratory campaign.The campaign reused settings and seeds from earlier work, so it is not an independent or held-out confirmation.Source: §2.3 · p. 3

2.4 Endpoint and accounting

• The finite-window endpoint charges road, entry-queue, and removed or abandoned trips over time per offer, including inherited backlog but not restoring removed vehicles’ physical effects.It is a network-burden measure, not delay above free flow, completed-journey time, loss-free capacity, or traveller welfare.Source: §2.4 · p. 4

2.5 Statistical analysis

• Inference uses ten seed-paired differences, Student intervals, and Holm-adjusted comparison families; choosing the best fixed plan after seeing results remains exploratory.The Student analysis assumes independent seed-level differences and approximate normality; the sign-flip sensitivity calculation still requires symmetry or valid within-pair exchangeability.Source: §2.5 · p. 4

3.1 The maximum-green setting changes the comparison

• At 0.24 arrivals per second, default actuation has a mean endpoint of 2198.1 s versus 1396.6 s for the 60 s fixed plan, a 57.4 per cent increase.Reducing only the maximum-green multiplier from 2 to 1.25 lowers the endpoint to 409.2 s, 70.7 per cent below the same reference; the multiplier-3 endpoint is 3846.3 s.Source: §3.1 · p. 5, Figure 1 · p. 5

3.2 Retiming the reference reduces the apparent benefit

• The 40 s fixed plan has the lowest mean endpoint among the three tested fixed settings at every load, but its best setting remains at the sweep boundary.At load 0.24, its endpoint is 473.1 s, compared with 1396.6 s at 60 s and 3232.7 s at 90 s.Source: §3.2 · p. 5
• Discharge-based termination has the lowest observed mean across all eight configurations, but its advantage is smaller against the retimed fixed baseline.All ten paired differences favor termination at each load, but the comparison does not isolate discharge sensing from other controller differences.Source: §3.2 · p. 5, Table 2 · p. 6

3.3 What the recorded traffic explains

• At the highest load, default actuation ends 96.8 per cent of greens on absence of demand and 3.2 per cent at maximum green, with an 8.8 s mean green.Network-wide counts cannot rule out consequential long greens at a small number of junctions.Source: §3.3 · p. 6
• The four lowest-mean configurations finish with about 88–108 vehicles on-road or waiting for entry, versus 308–792 for the other four; removals range from about five to 249 per hour.Because removal can bound occupancy while journeys remain unserved, these counts do not demonstrate stability.Source: §3.3 · p. 6

Start-up loss and termination settings

• Near-zero simulated start-up lost time may favor frequent phase changes, while a 4 s no-discharge rule can end green because of sparse arrivals or slow start-up as well as blockage.No direct blockage classification was recorded, so the termination benefit cannot be assigned entirely to spillback avoidance.Source: §4 · p. 6

Finite horizon and physical removals

• The common finite measurement window does not establish stationarity, and physical removals or entry abandonment need sensitivity analysis.A vehicle stationary for 400 s may be persistently queued rather than irreversibly deadlocked; charging lost journeys does not restore their physical effects.Source: §4 · p. 6

One model and one demand pattern

• One uncalibrated model and synthetic demand limit real-network validity, and shared assumptions do not protect relative rankings from omitted features.The study calls for matched replication in an established simulator followed by field calibration before deployment claims.Source: §4 · p. 7

Exploratory selection and uncertainty

• Ten paired seeds, post hoc selection of the 40 s plan, and sweeps that do not bracket an optimum make the reported reductions exploratory rather than gains over optimally tuned fixed-time control.The discharge threshold was not optimized, and the selected comparison lacks independent confirmation.Source: §4 · p. 7

5 Conclusion

• Changing maximum green reverses the comparison between presence-based actuation and the original fixed plan on this simulated urban network.A shorter fixed plan substantially reduces the apparent advantage of responsive control, while discharge-based termination retains a lower observed mean.Source: §5 · p. 7
• The methodological conclusion is to compare timing settings and retimed references before attributing performance to a controller family.The experiment does not benchmark all possible controller families.Source: §5 · p. 7
• A deployable setting requires evidence beyond this experiment about start-up loss, unresolved demand, other traffic patterns, and model validity.Source: §5 · p. 7

Data and code availability

• The source is available under the PolyForm Noncommercial 1.0.0 licence, and the analysed campaign includes a CSV of all 240 run summaries.The campaign is saved at results/ceiling-and-retiming/2026-08-01T20-49-55-567Z.Source: Data and code availability · p. 7
• The exact experimental source and runtime are not preserved in a complete versioned archive, limiting long-term reproduction.Source: Data and code availability · p. 7
• Running scripts/paper-appendix.ts with the campaign directory regenerates the numerical tables and timing-figure data from saved runs, not new traffic simulations.Source: Data and code availability · p. 7

A Model parameters

• Table 3 presents vehicle profiles, distinguishing parameters that are fixed across drivers from those scaled by sampled driver factors.Source: Table 3 · p. 9
• Vehicle length, minimum gap, and emergency deceleration do not vary between drivers.Source: Table 3 · p. 9
• Table 4 presents other principal model and campaign parameters, reports normal draws as mean and variance, and points to Table 1 for timing overrides.Source: Table 4 · p. 10

B Full results

• The paired tables compare against fixed-time and report nominal 95 per cent Student intervals, paired Student-test p values, and unadjusted exact sign-flip values.The intervals are not simultaneous; bold configurations pass Holm correction.Source: B Full results · p. 11
• Tables 5, 6, and 7 report mean endpoint and traffic accounting at 0.12, 0.2, and 0.24 arrivals per second, respectively.Source: Table 5 · p. 11, Table 6 · p. 11, Table 7 · p. 11
• Tables 8, 9, and 10 give paired endpoint differences at 0.12, 0.2, and 0.24 arrivals per second, with Holm adjustment retaining the campaign’s 21-comparison family.Source: Table 8 · p. 12, Table 9 · p. 12, Table 10 · p. 12
• Table 11 reports phase-controller telemetry averaged across seeds; fixed-time arms are omitted because they do not record these counters.Source: Table 11 · p. 12

• The checks are previously recorded evidence, not new runs or additional replications of the timing experiment.Source: C Recorded simulator checks · p. 13
• None of the checks compares the study network with field observations.Source: C Recorded simulator checks · p. 13
• Because synthetic scenarios share the main experiment’s model assumptions, their agreement indicates internal consistency rather than independent validation of field performance.Source: Table 12 · p. 13