Simulation methodology

1. Simulation core

The engine is a time-driven microscopic simulation: every vehicle is an individual agent advanced in fixed steps of 100 milliseconds (configurable) along the arc length of its lane. All vehicle dynamics, lane changes, gap acceptance, signal responses and route choices are resolved per step. The engine is a pure function of its inputs — the road network, the demand scenario and a random seed — and runs in a dedicated worker thread, decoupled from rendering.

Road Mesh does not embed a third-party simulation engine such as Eclipse SUMO or PTV Vissim; the models below are implemented directly and unit-tested against published formulations and reference behaviors.

2. Car following

Longitudinal control uses the Improved Intelligent Driver Model (IIDM) with the Constant-Acceleration Heuristic (CAH) as formulated by Treiber and Kesting. Each driver profile carries the standard parameter set from the car-following literature: desired time headway T, minimum standstill gap s0, maximum acceleration a, comfortable deceleration b, and acceleration exponent delta. The CAH term prevents the overreaction to cut-ins that the plain IDM exhibits, giving realistic behavior in dense merging traffic.

3. Lane changing and cooperation

Discretionary and mandatory lane changes use MOBIL (Minimizing Overall Braking Induced by Lane changes; Kesting, Treiber & Helbing), which accepts a lane change when the acceleration advantage outweighs the imposed braking on followers, weighted by a politeness factor, and never forces a follower below a safe-braking limit. On top of MOBIL the engine models cooperative yielding — vehicles open gaps for merging and weaving traffic — and zipper merging at lane drops.

Human factors are modeled per vehicle: bounded perception (an Ornstein–Uhlenbeck estimation-error process when driver awareness is below 1), discrete action steps rather than per-tick reactions, distribution of desired speeds, and per-vehicle parameter jitter so no two drivers are identical. Each vehicle draws from its own seeded random stream, so one vehicle's behavior never perturbs another's random draws.

4. Junctions, priority and signals

Unsignalized junctions resolve conflicts through explicit conflict-point yield roles (aware of left- and right-hand traffic) and gap acceptance with configurable critical gaps. Roundabouts implement entry yield to circulating traffic with an entry-capacity behavior calibrated against Highway Capacity Manual roundabout capacity relationships, including curvature-derived negotiation speeds and rolling entries.

Traffic signals are fixed-time controllers with phase-based protected and permissive movements, yellow and all-red intervals, coordinated offsets across multi-node junction areas, and optional imported timings (OCIT-C supply data or SUMO tlLogic). Permissive movements yield through gap acceptance during their phase.

5. Routing

Vehicles choose lanes freely along a route computed by per-destination reverse Dijkstra search over the lane connector graph. Turn restrictions, lane-use restrictions (such as bus lanes) and closures constrain the graph; optional congestion-responsive rerouting periodically re-weights turns by observed downstream queuing.

6. Determinism and reproducibility

Every run is exactly reproducible: the same network, scenario and random seed produce an identical simulation, tick for tick. Stochastic elements — destination choice, vehicle and driver profile assignment, desired-speed variation, parameter jitter and perception noise — draw from a seeded deterministic generator (mulberry32) with independent per-vehicle substreams. Changing the seed samples a different realization of the same demand; holding it fixed lets an edit's effect be isolated from random variation. A configurable warm-up period excludes the network-filling transient from summary statistics.

7. Reported metrics and Level of Service

Vehicle delay is measured as time lost versus the vehicle's own free-flow speed, integrated over its trip. For junction analysis, movement delay is scoped to the junction: it accrues from entry into the approach lane until the vehicle clears the junction, which makes it an estimate of HCM control delay. Level-of-Service grades A–F apply the HCM thresholds for signalized intersections (Exhibit 19-8) and unsignalized intersections and roundabouts (Exhibit 20-2) to this volume-weighted estimated control delay, with warm-up excluded. Because the delay window starts at approach-lane entry and no volume-to-capacity override is applied, grades are labeled as estimates throughout the application and its exports. Queues are counted in vehicles below a standstill speed threshold per lane.

References

• Treiber, M., Hennecke, A., & Helbing, D. (2000). Congested traffic states in empirical observations and microscopic simulations. Physical Review E, 62(2).
• Kesting, A., Treiber, M., & Helbing, D. (2007). General lane-changing model MOBIL for car-following models. Transportation Research Record, 1999(1).
• Treiber, M., & Kesting, A. (2013). Traffic Flow Dynamics: Data, Models and Simulation. Springer. (IIDM and CAH formulations.)
• Transportation Research Board. Highway Capacity Manual, 6th Edition. (Signalized, unsignalized and roundabout LOS criteria and capacity relationships.)