Car-Following Models#
This overview page collects information regarding car-following models in sumo. It is still a stub and needs be expanded.
Links#
- Publications on car-following models
- How to implement a new car-following model
- More implementation notes on car-following models
Model Parameters#
Most car-following models have their own set of parameters. Because some models have a common ancestry they share some parameters. These are described below.
Reaction time#
By default, drivers may react to their surroundings in every simulation step (i.e. by changing their speed or changing lanes). Thus, their time to react, is equal to the value set by sumo-option --step-length (default 1s). To change the frequency of decision making and thereby delay reaction, the parameter actionStepLength may be set.
actionStepLength#
This parameter is used to decouple the simulation step length (option --step-length)
from the frequency of driver decision making. It can be set either in
the <vType>-definition as attribute actionStepLength or by setting a default
value for all vTypes using option --default.action-step-length <FLOAT>.
By default, the action step length is equal to the simulation step length which works well for the default step length of 1s. When performing sub-second simulation by setting a lower step-length value, it may be useful to maintain a higher action step length in order to model reaction times and also in order to reduce computational demand and thus speed up the simulation.
Note
The action step length works similar to a reaction time (vehicle will not react immediately to changes in their environment) but it also differs from a "true" reaction time because whenever a vehicle has it's action step it reacts to the state in the previous simulation step rather than to the state that was seen in their previous action step. Thus the Perception-Reaction loop is less frequent but still as fast as the simulation step length.
tau#
This parameter is intended to model a drivers desired minimum time headway (in seconds). It is used by all models. Drivers attempt to maintain a minimum time gap of tau between the rear bumper of their leader and their own (front-bumper + minGap) to assure the possibility to brake in time when their leader starts braking.
The use of minGap with respect to the desired (minimum) time headway varies by model. I.e. the Krauss model selects a speed that ensures minGap can always be maintaned whereas other models may not do this.
As explained above, the reaction time is defined by the simulation step length and the actionStepLength parameter. fundamentally limited by the simulation step size (sumo option --step-length <FLOAT>).
By setting a value of tau that is higher than the reaction time, the driving saftey can be increased since vehicles will use large gaps and have enough time to react. Conversely, setting a value of tau that is lower than the reaction time leads to unsafe driving. In the simulation this can manifest as high deceleration or even collisions.