Zipper Merge Simulator

Car following. Every vehicle uses the Intelligent Driver Model (IDM, Treiber et al. 2000), the standard research model: acceleration a·[1 − (v/v₀)⁴ − (s*/s)²] with desired gap s* = s₀ + vT + vΔv / (2√(ab)). Cars: a = 1.4 m/s², b = 2.0 m/s², T = 1.3 s, s₀ = 2 m, 4.5 m long. Trucks: a = 0.7, b = 1.5, T = 1.8 s, 14 m long. These are the same for every strategy, so differences come only from merging behaviour. Time step 0.1 s.

Lane changes. A car can change lanes in two ways. (1) Into a genuinely free gap: the car behind would not have to brake, and the move doesn't take space that car is about to drive into. (2) In front of a driver who is deliberately letting it in. That driver brakes at up to 3 m/s² (4 m/s² for a car that has waited at the end for more than 20 s), and the merger accepts tighter gaps the closer it is to the end of its lane. Drivers who yield brake at up to 3 m/s² to open a gap.

Throughput = vehicles crossing the merge point during the measured window ÷ window length. Delay for one vehicle = its actual trip time − the time it would take at its own desired speed. It is accumulated every step as (1 − v/v₀)·Δt, and time spent waiting to enter the road (when the queue reaches the start) counts in full. Total delay (vehicle-hours) is the sum over everyone during the window. Average delay covers every vehicle that arrived in the window; trips not yet finished count with the delay they have built up so far, so a growing queue is never hidden. Queue length = length of road (per lane) occupied by slow traffic (< 10 mph). Slow cars less than ~330 ft apart count as one queue, so moving cars inside a stop-and-go wave don't split it. Cars waiting to enter add about 25 ft each. Backup reach = how far upstream of the merge point the slow traffic starts, which is what an arriving driver sees.

Warm-up. Nothing is measured during the warm-up, so the road starts out realistically loaded.

Fair comparisons. The arrival stream (times, lanes, vehicle type, desired speed, and each driver's random strategy draw) comes from its own seeded random generator, independent of what happens on the road. "Compare strategies" therefore runs every strategy on exactly the same traffic.