Session 55_1 Andreas Tapani

En vidareutveckling av the Intelligent Driver Model

Johan Olstam och Andreas Tapani

Contents

� The intelligent Driver Model (IDM)

� The Human Driver Model

� Problems with the IDM

� A modified version of the IDM

� Microscopic comparison

� Macroscopic comparison

� Conclusions

The Intelligent Driver Model (Treiber et. al 2000)

( ) ( )

( ) ( )

( )( )

( )

4

0

2

0

00

, ,

1

,,

,2

free int

freedes

int

des

a a v a s v v

va v a

v

sa s s a

s v v

v vs v v s vT

a b

-*

*

*

= + D

æ ö÷ç= - ÷ç ÷çè ø

æ ö÷ç= - ÷ç ÷è øD

DD = + +

s

Advantages with the IDM

The IDM has

� reached good scores in model cross-comparison tests

� physically interpretable parameters such as desired speed (vdes), desired time gap (Tdes),

and desired maximum acceleration (a0) and

deceleration rate (b)

The Human Driver Model (HDM)

(Treiber et. al 2006)

The HDM includes

� finite reaction times (T’)

� spatial anticipation (considering n>1 leaders)

� temporal anticipation (anticipating the change in relative position and speed to the considered leaders)

� modeling of drivers' imperfect estimation capabilities of the surrounding vehicles' position, speed, etc.

A version of the HDM extended IDM was in Olstam et. al. (2009) used within a model for simulating surrounding vehicles in a driving simulator.

( ) ( ) ( )1i

free inti iji

j i n

a t a t T a t T

-

= -

¢ ¢= - + -å

Observed problems with the IDM

The simulated vehicles did not reach their assigned

desired speed due to negative interaction acceleration

(aint

) even if the distance to a preceding vehicle (s) is

longer than the estimated desired following distance (s*)

0 1 2 3 4 5−5

−4

−3

−2

−1

0

1

2

s/s*

ain

t /a0

0

4

2 0.25

0 if 0.75des

intss

v

v

a a

a

* = Þ = -

Þ ³ <

Simulated vehicles stays in the left lane…

Vehicle B does not change to the right lane since the

IDM will induce a deceleration even if vehicle A is far

away and drive at the same speed as B

The modified IDM

( )

( ) ( )

2

3 20 1 3 2 3

1

1

0

s ss s

int s s ss s s

ss

a a c c c c b

b

* *

* * *

*

-ìï <ïïïï= - × - + - £ £íïïïï >ïî

( ) ( )

( )

( )

1 23 2

32

3 4 2

2 4 2 5, ,

1 1

1 5 1 2 5,

3 2 27 2

c c

c c

b b

b b

b b b

b b

- -= - =

- -

+ - -= = -

- -

0 1 2 3 4 5−5

−4

−3

−2

−1

0

1

2

s/s*

ain

t /a0

original IDM model

modified IDM model

Microscopic comparison

Comparison of the two models with the Bosch

urban trajectory data set has been done

� Comparable results (although manual

calibration of the modified IDM)

� The modified model gives oscillating

acceleration trajectories (has been observed in

real traffic and it is the basis of the Wiedemannand Fritzsche models)

Speed-flow comparison

0 500 1000 1500 2000 2500 3000 3500 4000 4500 50000

10

20

30

40

50

60

70

80

90

100

110

120

130

140

150

Flow [vehicles/h]

Space m

ean s

peed [km

/h]

Original IDM (black)

Modified IDM (grey)

SRA speed−flow relationship

0 500 1000 1500 2000 2500 3000 3500 4000 4500 50000

10

20

30

40

50

60

70

80

90

100

110

120

130

140

150

Flow [vehicles/h]

Space m

ean s

peed [km

/h]

Original HDM/IDM (black)

Modified HDM/IDM (grey)

SRA speed−flow relationship

IDM HDM/IDM

Conclusions

� The modified IDM result in a higher average speed for a specific flow level, a less steep speed-flow relationship and higher capacity.

� The modified IDM show an improved agreement with the speed-flow relationships in real traffic.

� The IDM's good ability to reproduce vehicle trajectories seems to be retained in the modified IDM.

� Modeling spatial anticipation makes it more important to capture the car-following acceleration behavior compared to if only one leader is considered.

� The importance of car-following and lane-changing models being well integrated is illustrated by the problem observed with the IDM model in connection with lane changes.