Back to the experimentTHE EVIDENCE BEHIND THE EXPERIENCE

The traffic ripple: sources & model

An educational adaptation of the Intelligent Driver Model (IDM), with a manual braking experiment on a closed road.

Scientific review · independent subject review pending

The source and model records are available for inspection. No external scientific reviewer has signed off yet.

traffic-idm-1 · replay format 1

What supports the explanation?

A slowdown can grow through interactions between drivers, without a road obstruction.

MIT explains upstream-moving queues; the research team documents real ring-road experiments.

MIT Engineering

Desired speed, current gap, relative speed, and preferred time headway influence IDM acceleration.

The model authors provide the acceleration equation, desired-gap equation, and meanings of the parameters.

IDM model authors

Real vehicles have produced stop-and-go waves on experimental ring roads.

These field experiments provide evidence for the phenomenon; they do not validate every setting in this playground.

Phantom Traffic Jams research team

What this model assumes

A 600 m, one-lane ring; identical 4.5 m vehicles and driver parameters. The rendered road and cars are illustrative, not a scale drawing.
12–52 cars; desired speed 30–90 km/h; desired time headway 0.6–2.4 s. These are learning controls, not recommended driving settings.
Acceleration parameter a = 0.7 m/s², comfortable deceleration b = 1.8 m/s², minimum desired gap 2 m, speed exponent 4. These choices are not calibrated to a particular road.
Manual braking lasts 2.5 simulated seconds and applies a −4.5 m/s² limit. The application also caps deceleration at −7 m/s² and adds a 0.2 m spacing limiter. These are application additions to IDM.
Fixed 1/60 s time steps; nonnegative speed. The spacing limiter is a numerical guard, not a proof that all possible configurations are collision-free.
“Stopped or crawling” counts speed below 0.5 m/s (1.8 km/h). Time headway is a spacing preference, not an explicit reaction delay.
No lane changes, intersections, driver variation, or realistic crash dynamics. Each experiment is limited to 10 simulated minutes and 200 brake actions.

What has been checked

Deterministic replay, bounded input validation, selected numerical invariants, and unit conversions are covered by automated tests. These checks do not establish real-road predictive accuracy.

Each source supports the associated claim. Sources do not certify this implementation or its visuals.

Our review process