Microscopic Modelling of Driver Behaviour in the presence of Other Moving Vehicles

Introduction:

A microscopic theory of traffic flow, i.e.,a theory of traffic flow which can predict the action of individual drivers over time in different driving situations, is an important and essential aspect of the area of traffic science and engineering. Since the early 1990s, microscopic simulation models have emerged as important tools for intelligent transportation system strategies.As basic and important components of a microscopic traffic simulation model, the car-following models has received considerable research interest.

Traditionally, traffic models have been developed using fluid flow analogies. This assumes that a highway can be represented by a link that has a certain flow capacity. This may be acceptable when traffic flows freely but this model modelling approach breaks down under congested traffic conditions.

Microscopic models differ from traditional computational models in that they have significant random components and attempt to represent the dynamic nature of traffic systems by simulating the interactions of individual vehicles. They model numerous components of traffic flow and their influences, and therefore provide more accurate traffic flow information, which is necessary for the analysis of congested road networks.
Microsimulation models also differ in that vehicles move in the system in real-time and are modelled according to the behaviour of vehicle drivers.
Researchers have indicated that vehicle drivers perform according to their aggression and awareness.In traffic engineering terms this can be accounted for in terms of simple rules of car-following, gap acceptance and vehicle kinematics.

These models can be developed using various approaches viz fuzzy inference based approach, multiregime approach etc.In the following work, a structure based on the potential field theory of robot motion planning which can be used to model driver behaviour in variety of driving situations is being discussed.

Modelling Driver Behaviour

Traditionally driver behaviour has been modelled  differently for different driving scenarios. For example there is a model for car-following behaviour, a model for lateral displacement etc.A single model structure that can predict driver behaviour in various driving situations will be discussed.

The driver's  behaviour at all times, it is assumed, is motivated by his concern for safety and his urge to minimize the travel time. Further it is assumed that the driver views his travelling environment as a conglomeration of obstacles which need be to be avoided in order to proceed safely.
Sometimes the driver may choose a point on the road as a goal point which he would like to reach while driving in this environment of obstacles. Such an understanding of the driving allows the development of a single model structure because all features such as road-edge, pot-hole, lane marking, other vehicles on the road, etc can be viewed as obstacles albeit with different properties.

The model being discussed will be able to predict the driver's (or the test vehicle's ) action in terms of choice of steering angle and acceleration or
deceleration , given a driving scenario. First the principles on which the model is based will be discussed followed by the description of the model to predict the position of the test vehicle on the road and the velocity of test vehicle over time. 

Principles of the proposed model

1.)  The driving environment is characterized by the presence of obstacles , either static or dynamic, and in certain cases the presence of an 
       immediate goal.
2.)  Each obstacle emanates some potential (or force ) around it; the potential (or force)  field of an obstacle is repulsive in nature. A goal also 
      emanates a potential field , which however is attractive in nature. 
3.)  The potential field emanated by an obstacle depends on the properties of the obstacle. For example, the potential field due to a parked   
       vehicle may be assumed to be more prominent than the potential field due to a small pothole.

4.)  The potential at any point on the road is the algebraic sum of  all the potentials due to the driving environment features (i.e.,obstacles and  
       goal) around that time.

5.)  The test vehicle is assumed to move from one cross section of the road to the next. These cross-sections can be arbitrarily close.

6.)  While moving from one cross section to the next, the driver of the test vehicle always chooses a point with smallest potential among  all the 
      accessible points depend on the test vehicles present orientation and the maximum steering angle.

7.)  There exists an inverse relation between velocity at a point and the potential at that point. The velocity corresponding to the potential at  a  
       point is referred to as the sustainable velocity. For example, the speed on a narrow empty road is generally lower than the speed on a   
       broad empty road.  

8.)  The response of the test vehicle at a given point in terms of acceleration or deceleration (referred to as longitudinal response as opposed to
       lateral response) is a function of the rate of change of potential ( faced by the driver) as well as the difference between the current velocity 
       of the test  vehicle and the sustainable velocity at that point.

Description of the model

A model of driver behaviour is developed based on the principles presented above.

Potential  computation

Let (x,y) represent the Cartesian space in which the road as well as all its features are defined; let Ui(x,y) be the potential field emanated by obstacle i;and  let U(x,y) be the potential at the point (x,y) on the road;

Then based on principle 4, the potential at any point (x,y) is calculated as  U(x,y) =   Si Ui(x,y)

Position  profile  determination  model

The term "position profile" refers to the path of of  TV  on the road over time as predicted by the proposed model. In general, the position profile is described graphically with distance from the left edge of the road as the abscissa and distance along the road as ordinate.

Position profile as suggested in principle 6 , is affected not only by the potential at different road points but also by the constraint of the maximum steering angle. This latter constraint restricts the number of points which the TV  can access from its present location.

If ( xn,yn)  represents the coordinates of the nth cross section and  (xna,yna)  represents the coordinates of the accessible points in the  nth cross section then as per principle 6 , the position of the TV (xn*,yn*) , at the nth cross section is obtained such that 
                     U (xn*,yn*) = min  U(xn,yn) ,                (xn,yn) € An  

  where An  represents the set of all accessible points in the nth cross section.

The set An  can be determined from the current orientation of the TV and the maximum steering angle restrictions.

Velocity profile determination model 

The " velocity profile" refers to the predicted velocity of TV along the position profile. The predicted velocity is obtained from the acceleration or decelerations (longitudinal response) of the TV at every instant of time . The acceleration/ deceleration is obtained using the longitudinal response model developed on the basis of principle 8.

If a(t + T) is the response of the TV at time t + T then as per Principle 8: 
             a(t+ T) = f (U'(x*,y*) ,Vs(t) , Va(t))  

where , T is the reaction time of driver , (x*,y*) is the position of the vehicle at time t , U'(x*,y*)  is the rate of change of potential faced by TV at time t, Va(t)is teh actual velocity of TV at time t, Vs(t) is the sustainable velocity of teh TV at time t.The sustainable velocity is an inverse function of U(x*,y*)  ; i.e., Vs(t) = g(U(x*,y*) ). In this work the functional form used is  Vs(t) = c1-c2 U(x*,y*)  where c1 and c2 are positive constants.

Functional form for longitudinal response model 

Proposed Potential functions

As stated earlier, the model views the driving environment as a set of obstacles and immediate goals (jointly referred as features) . Further, the model also assumes that these features emit a potential field and that drivers react to the potential they face while travelling.

Goals are assumed to be of a single kind which basically states the preferred location of the driver of TV. However ,obstacles can be of various kinds. The two broad classes of obstacles are: (i) static obstacles (ii) dynamic obstacles . Static obstacles include features like road edges, pot-holes, lane marking , parked vehicle etc. whereas dynamic obstacles are primarily of two types : (i) vehicle moving in the direction of TV  (referred to as same direction vehicle, SDV) (ii) vehicles moving in the opposite direction of TV ( referred to as opposite direction vehicles , ODV) .

In the following discussion potential field functions for dynamic obstacles and goals only would be presented

Potential function for SDV

Desirable properties 

1.) The potential field function at all points should be greater than or equal to zero indicating repulsion.

2.) The potential function should be continuous and should decrease monotonically as one moves away from the SDV.

3.) The spread of the potential field in the longitudinal direction behind the SDV is more than that ahead of the SDV.

4.) Potential field should be more prominent in the longitudinal direction than in the lateral direction. The implication of this is that the field in the lateral  direction assumes large values only in a direction  very close to the SDV.

Representation of Potential function 

                        max [0, h(x,y)]              (x,y) lies in any zone
       USDV =     min [h(x,y), h1(y)]         h(x,y) >= 0 and  (x,y) lies in  Zone-III , Zone-IV
                        min [h(x,y), h2(y)]         h(x,y) >= 0 and  (x,y) lies in  Zone-V , Zone-VI

 where,
             h(x,y) = [ s1/|x -x0| + s2]d  +  [s3 (lid - | y - y0|)(1- |x -x0| /s4)](1-d )
             h1(y)  = US [1 - |y -y0|/(lid1 + lid2) ]
             h2(y)  =   US (1 - |y -y0|/ lid3  )

where, (x0,y0 )  are the co-ordinates of SDV, US is the potential at  (x0,y0 ) , lid1 , lid2 and lid3 are the distances from the SDV in the y-direction.
 In the above equation values of s1, s2 , s3 , and d depend on the co-ordinate of points (x,y) and the zone in which (x,y) lies on the road.

The above function although complex in its algebraic description , produces a potential field function similar to the one described before  

               

Potential function for ODV

Desirable properties 

1.) The potential field function at all points should be greater than or equal to zero indicating repulsion.

2.) The potential function should be continuous and should decrease monotonically as one goes away from the ODV.

3.) The spread of the potential field in the longitudinal direction ahead of the ODV is more than that behind  the ODV.

4.) Potential field should be more prominent in the longitudinal direction than in the lateral direction. The implication of this is that the field
     in the lateral direction assumes large values only in a direction  very close to the ODV.

5.) The longitudinal spread of potential field for ODV in front of ODV should be more than the spread of potential field (for SDV) behind the SDV.

6.) The lateral spread of potential field (for ODV) in front of ODV  should be more than the spread of potential field (for SDV) behind SDV.

Representation of potential function

 The potential function is of the following functional form (x0,y0 )

                 max [0,g(x,y)]              (x,y) lies in any Zone
UODV =    min [g(x,y) , g1(y)]       g(x,y) >= 0 and (x, y) lies in Zone -I
                 min[g(x,y) , g2(y)]       g(x,y) >= 0 and (x,y) lies in Zone -II

where,
          g(x,y) = s01/|x -x0| +  s02
            
g1(y) = U0 (1- |y-y0|/lod1)
         g2 (y) = U0(1- |y-y0|/lod2)

where , (x0,y0 ) are the coordinates of ODV, Uis the 

 

Potential Function for the goal

Desirable properties 

1.) The potential field function at all points should be less than or equal to zero indicating attraction.

2.) The goal potential function should be continuous.

3.) The absolute value of the potential field function should increase with lateral deviation. 

4.) The effect of lateral deviation on the potential field function should reduce with longitudinal distance from the goal. 

5.) The influence area of a goal depends on the context in which it is used . This is so, because under different scenarios, the effect of goal may be  
     felt differently.

Representation of the potential function 

The following functional form of the potential is developed 

                max [f(x),f1(y)]      f(x) <= 0  and (x,y) lies in Zone -I
UGoal =    max [f(x) ,q]         f(x) <= 0  and (x,y) lies in Zone -II
                max [f(x),f2(y)]     f(x) <= 0 and  (x,y) lies in Zone -III

where ,  f(x) = -|x -xg| / r
             f1(y) = q [1 - (y-yg + lg2)2 /l2g1 ]1/2 
             f2(y) =  q[ 1- (y - yg)2 /l2g3 ]1/2

Implementation and Results

Two different scenarios namely, unidirectional and bi-directional flow are studied. Under the unidirectional flow, driver behaviour in car-following and overtaking  situations are studied. Further, in car-following situations the analysis of local stability and asymptotic stability have been carried out.
Under bi-directional flow, driver behaviour for various road widths ( from narrow to standard road widths ) have been studied.

Parameter values for the potential field function of SDV, ODV, and goal are obtained from trial and error. Parameter values for  driver response both for longitudinal as well as lateral model are obtained from real- world data .

Road edges are also obstacles, albeit static, which contribute to the potential value at any point on the road. The potential field function is as stated: ULRE = a1 exp(-b1x) + a2 exp[-b2(w-x)]

Car-following situation

In order to obtain the results for car-following situation a restriction on lateral movement of vehicles is used since a vehicle will follow whenever lateral movements will not allow an eventual overtaking . the analysis of car following helps in evaluating various properties of car-following namely, local stability, asymptotic stability, independence of stable condition from initial velocity , actions of leading vehicle , dependence on final velocity and closing-in and shying away. Under the car-following situation ,the following potential field functions are active in the traffic environment:

1.) Road edges
2.) SDV (which acts as the leading vehicle, LV)

Analysis of local stability

Local stability analysis leads with how the perturbations to the following vehicle's (FV) actions introduced by LV propagates with time.As the time increases they should decrease and eventually come to zero,implying a stable condition.The model produces actions of FV which are locally stable.

Independence of stable condition from initial distance headway

The stable condition is independent of the initial distance headway between the vehicle.The stable distance headway is same for all the sub-scenarios within a particular scenario.This implies that the stable distance headway in particular and the stable condition in general, are independent of the initial distance headway. 

Independence of stable condition from initial velocity

FV in all sub-scenarios stabilizes at the same distance headway irrespective of the initial velocity of LV an FV implying the independence of stable condition from initial velocity. 

Independence of stable conditions from actions of LV 

FV in all sub-scenarios stabilizes at the same distance headway irrespective of the actions of the LV. 

Dependence of stable condition only on the final velocity

 

Analysis of asymptotic stability

Asymptotic stability deals with how the perturbation introduced by the LV propagates in a platoon of vehicles following the LV.A platoon is described as asymptotically stable if the perturbatios introduced by the LV reduces as it is transmitted upstream in the platoon.

The model described here conserves the property of asymptotic stability in car following situations.

Conclusion

The model works well as we can see , in car following situations for it conserves the property of local and asymptotic stability.

References

1.) Preliminary exploration into driver behavior , M.Tech thesis, Saurabh Agarwal, Deptt of Civil Engg
2.) Modelling driver behaviour in the presence of  other moving vehicles, Vasishtha, M.Tech Thesis, Deptt of civil Engg, IITK.
3.) A Behavioral car- following model for computer simulation,P.G Gipps.

Thank You.