A Byte of Wall StreetFrom the video: The Fast Lane Is a Trap in Traffic
Interactive · which lane is fastest?

Drive the jam yourself, then check the clock

Pick a bottleneck and a lane strategy. Your car drives a 3.7-mile freeway and is timed against ordinary drivers who entered the same minute. What 4,404 real cars did is further down.

your car prefers the left lane other drivers tap the strip to look elsewhere

Lane speeds · timed stretch

Your car

Waiting for your car to reach the road…
Real data · not the simulator

On US-101, the left lane was the slowest place to sit

NGSIM camera-tracked vehicle trajectories, Los Angeles, 15 June 2005, 7:50–8:35 a.m., about 1,800 ft of freeway, 4,404 cars, average pace about 21 mph. Cars that stayed in one main lane, seconds per mile against same-minute traffic (negative is faster):

  • The far-right main lane was the fastest main lane in 9 of 9 five-minute windows; the left lane was the slowest in 7 of 9.
  • The left lane held about 15% more vehicle-time than the right lane and ran about 17% slower.
  • Changing lanes barely paid: cars that used three or more lanes were 2.6 s/mile faster than same-minute traffic, cars that never changed were 2.0 s/mile slower.

One freeway, one morning, and drivers chose their own lanes, so this describes what happened rather than proving why. The simulator above is for seeing the mechanisms; quote these numbers, not its.

How the simulator works

Small rules, every car

  • Each car follows the Intelligent Driver Model (Treiber et al., 2000): it keeps a safe time gap and brakes smoothly. Desired speeds average about 69 mph.
  • Lane changes use MOBIL (Kesting et al., 2007): change if it helps you more than it costs the drivers behind. Drivers also chase a lane that looks faster ahead, accept a short gap right after changing, and get let in when they signal for a forced merge.
  • One car in ten is a test driver: equal numbers stay left, middle, right, or weave, and they drive the same way otherwise. You ride one of them.
  • Each test car's time over the timed 2.5 miles is compared with ordinary cars that entered within a minute of it, the same method used on the real data.

It is a port of the Python model from the video, checked against field data for lane speeds in light traffic and the speed of stop-and-go waves. Real traffic has more going on, so treat its numbers as direction, not size.