Intelligent Driver Model

In traffic flow modeling, the intelligent driver model (IDM) is a time-continuous car-following model for the simulation of freeway and urban traffic. It was developed by Treiber, Hennecke and Helbing in 2000 to improve upon results provided with other "intelligent" driver models such as Gipps' model, which loses realistic properties in the deterministic limit.

Model definition

As a car-following model, the IDM describes the dynamics of the positions and velocities of single vehicles. For vehicle $\alpha$, $x_\alpha$ denotes its position at time $t$, and $v_\alpha$ its velocity. Furthermore, $l_\alpha$ gives the length of the vehicle. To simplify notation, we define the ''net distance'' $s_\alpha := x_ - x_\alpha - l_$, where $\alpha - 1$ refers to the vehicle directly in front of vehicle $\alpha$, and the velocity difference, or ''approaching rate'', $\Delta v_\alpha := v_\alpha - v_$. For a simplified version of the model, the dynamics of vehicle $\alpha$ are then described by the following two ordinary differential equations:

:$\dot_\alpha = \frac = v_\alpha$
:$\dot_\alpha = \frac = a\,\left( 1 - \left(\frac\right)^\delta - \left(\frac\right)^2 \right)$
:$\texts^*(v_\alpha,\Delta v_\alpha) = s_0 + v_\alpha\,T + \frac$

$v_0$, $s_0$, $T$, $a$, and $b$ are model parameters which have the following meaning:
* ''desired velocity'' $v_0$: the velocity the vehicle would drive at in free traffic
* ''minimum spacing'' $s_0$: a minimum desired net distance. A car can't move if the distance from the car in the front is not at least $s_0$
* ''desired time headway'' $T$: the minimum possible time to the vehicle in front
* ''acceleration'' $a$: the maximum vehicle acceleration
* ''comfortable braking deceleration'' $b$: a positive number
The exponent $\delta$ is usually set to 4.

Model characteristics

The acceleration of vehicle $\alpha$ can be separated into a ''free road term'' and an ''interaction term'':

:$\dot^\text_\alpha = a\,\left( 1 - \left(\frac\right)^\delta \right) \qquad\dot^\text_\alpha = -a\,\left(\frac\right)^2 = -a\,\left(\frac + \frac\right)^2$

* ''Free road behavior:'' On a free road, the distance to the leading vehicle $s_\alpha$ is large and the vehicle's acceleration is dominated by the free road term, which is approximately equal to $a$ for low velocities and vanishes as $v_\alpha$ approaches $v_0$. Therefore, a single vehicle on a free road will asymptotically approach its desired velocity $v_0$.
* ''Behavior at high approaching rates:'' For large velocity differences, the interaction term is governed by $-a\,(v_\alpha\,\Delta v_\alpha)^2\,/\,(2\,\sqrt\,s_\alpha)^2 = -(v_\alpha\,\Delta v_\alpha)^2\,/\,(4\,b\,s_\alpha^2)$.

This leads to a driving behavior that compensates velocity differences while trying not to brake much harder than the comfortable braking deceleration $b$.

* ''Behavior at small net distances:'' For negligible velocity differences and small net distances, the interaction term is approximately equal to $-a\,(s_0 + v_\alpha\,T)^2\,/\,s_\alpha^2$, which resembles a simple repulsive force such that small net distances are quickly enlarged towards an equilibrium net distance.

Solution example

Let's assume a ring road with 50 vehicles. Then, vehicle 1 will follow vehicle 50. Initial speeds are given and since all vehicles are considered equal, vector ODEs are further simplified to:

:$\dot = \frac = v$
:$\dot = \frac = a\,\left( 1 - \left(\frac\right)^\delta - \left(\frac\right)^2 \right)$
:$\texts^*(v,\Delta v) = s_0 + v\,T + \frac$

For this example, the following values are given for the equation's parameters, in line with the original calibrated model.

The two ordinary differential equations are solved using Runge–Kutta methods of orders 1, 3, and 5 with the same time step, to show the effects of computational accuracy in the results.

This comparison shows that the IDM does not show extremely irrealistic properties such as negative velocities or vehicles sharing the same space even for from a low order method such as with the Euler's method (RK1). However, traffic wave propagation is not as accurately represented as in the higher order methods, RK3 and RK 5. These last two methods show no significant differences, which lead to conclude that a solution for IDM reaches acceptable results from RK3 upwards and no additional computational requirements would be needed. Nonetheless, when introducing heterogeneous vehicles and both jam distance parameters, this observation could not suffice.

See also

* Gipps' model
* Newell's car-following model
* Microscopic traffic flow model
* List of Runge–Kutta methods
* Traffic simulation

References

{{Reflist

External links

Interactive JS & HTML5 implementation of the intelligent driver model showing signalized intersections


* ttp://www.traffic-simulation.de/roundabout.html Interactive JS & HTML5 implementation showing the effect of different traffic rules at roundabouts
Common values of the IDM parameters and hints for the simulation

Textbook on traffic flow dynamics with a free downloadable chapter on the IDM

Road traffic management