Proceedings of the Fifth International Conference on Machine Learning and Cybernetics, Dalian, 13-16 August 2006 1-4244-0060-0/06/$20.00 ©2006 IEEE 911 REVIEW OF THE SIMULATION MODEL OF DRIVING BEHAVIOR XIAO-YUAN WANG, XIN-YUE YANG, GANG SHAN, FENG-QUN WANG Institute of Intelligent Transportation, School of Transportation and Vehicle Engineering, Shandong University of Technology, Zibo 255000, China E-MAIL: wangxiaoyuan@sdut.edu.cn, zbxinyue@gmail.com, sdshangang@gmail.com, wangfengqun@gmail.com Abstract: Study of driving behavior is the important content of the Intelligent Transportation Systems and the theoretic basis of the microscopic traffic flow simulation. In the past, the research concerning to driving behavior mainly focused on the car-following process. From the view of the behavior analysis, the car-following models concerning to the human factors based on cybernetics are reviewed in this paper, such as the GHR model, the collision avoidance model, the AP model, the fuzzy logic model, the neural network model, and the desired headway model, and so on. During the establishment, the advantage and the disadvantage are discussed in detail. Some important questions used to characterize the driving behavior, which are usually ignored in the establishment process of the models, are also revealed. With the latest research of the modern traffic field, the development tendency of the driving behavior simulation models in the future is forecasted from the specialization of the application field, the diversification of the study means and the practicability of the model establishment. Keywords: Driving behavior Microscopic simulation Car-following Modern traffic flow theory Intelligent transportation systems 1. Introduction Study of driving behavior is the key content of the Intelligent Transportation Systems (ITS) and the theoretic basis of the microscopic traffic flow simulation. In the past, researches are mainly focused on the car-following driving behavior. In recent years, study of the car-following driving behavior has become important in traffic engineering and security. Moreover, it has become the key content of the self-adapted cruise control strategy for developing the automatic road system and the cornerstone in a great deal of important traffic study area, such as (1) simulation modeling. Where, the car-following model controls the motion of the vehicles in the network. (2) the functional definition of advanced vehicle control and security systems, which are being introduced as a driver safety aid in an effort to mimic driver behavior but remove human error[1]. With the development of traffic science and technology, as well as the challenge of information burden, it is widely considered that the driving factor is the key to the development of the ITS. The research on car-following driving behavior is important for studying the traffic flow microscopic simulation, the driver guidance systems and the autonomous cruise control[2][3]. Compared with the existed models, the current driving behavior simulation models involve more human factors, which can reflect the real traffic situation better. Moreover, it is an important component of modern traffic flow theory, which can understand and expound the interaction between the micro-simulation and the macro-simulation. With the improvement of research methods and data collection means, the driver behavior simulation model is used as the theoretic basis for simulating and realizing the ITS. What is more, it has being the study hotspot in traffic flow theory and amount of effort that traffic engineers and psychologists have devoted to. In the future, based on the particularity of content and diversification of study means, the model study will develop into professional application, and many driving behavior simulation models will be designed to resolve some special problems. 2. Review In the past, researches concerning to the driving behavior mainly focused on the car-following process. Most of the car-following models for describing the driving behavior were developed on the basis of the cybernetics. The establishment theory can be described as follows: (1) drivers aim for optimal performance (2) driving is equivalent to the continuous application of a single control law (3) drivers use inputs that they may not be able to perceive but are somehow able to compute. For example, longitudinal distance, absolute speed of the leading vehicle and the acceleration of other vehicles in visual field, and so on (4) all the questions that cannot be explained by the model is noise that can be attributed to perceptual and control limitations.

Proceedings of the Fifth International Conference on Machine Learning and Cybernetics, Dalian, 13-16 August 2006 912 2.1. Gazis-Herman-Rothery Model (GHR) The study of Reuschel, A(1950) and Pipes, L.A(1953) shows that the analysis of car-following model was determined. It stands for the beginning of the research on the car-following model theory by analytic method. In 1958, Gazis, Herman and Rothery did a further study on the relationship between the micro-simulation model and the macro-simulation model and then, deduced the general formula of the car-following model based on the stimulus-response[4]. The model assumed that a vehicle can accelerate or decelerate by perceiving the stimulus between itself and the adjacent leading vehicle. Its formula is: [xn(t) (xn(t) t))m t) xn+1(t −xn+1(t)]   α(l,m)(xn+1(t       −xn+1(t))l ∆ + = ∆ + & & & & & (1) Where, (t) xn , (t) xn & , (t) xn & & are the position, speed, and acceleration/deceleration of vehicle n at time t t ∆ is the perception reaction time l and m are exponents (l, m) α is a constant whose dimensions are dependent on l , m The item inside the braces is sensitivity, and the item within the brackets is stimulus. The dominant properties of the model are described as follows[5]~[9]: (1) Local stability and asymptotic stability can be achieved. (2) No matter using test vehicle or film, there are errors in collection of experiment data, the model parameters and the optimal m, l cannot be calibrated accurately. (3) The model has no limit on the acceleration and deceleration which, in reality, is dictated by the dynamic performance of vehicle. Considering the fact that in real world unstable conditions do not last long but instead tend to settle quickly, limiting values of acceleration and deceleration should be incorporated. (4) It is a deterministic stimulus-response model. It assumes that the stimulus as well as the distance headway can be perceived precisely and that the driver can tune the response precisely. Due to the deterministic properties of the model, the inherent uncertainty in car-following process and the approximate essences of car-following behavior both are neglected. The stimulus-reaction properties are so simple that the complexity of psycho-physics behavior, such as stimulation, perception, reaction, and so on, can hardly be reflected when driving. (5) The response of the driver is based on only one stimulus, namely, the relative speed. Once the relative speed is zero the following vehicle neither accelerates nor decelerates irrespective of the distance headway between the vehicles. Owing to the single stimulus nature of the model, it fails to explain behavior such as closing-in and shying-away. 2.2. Car-following model based on safety distance Since 1970’s, there have been a great deal of researches on car-following models. Among the rest, the car-following model based on safety distance is, and should be the most fruitful and essential. Two approaches are mostly used to ensure the safety distance: (1) vehicle considers emergency braking of their leaders to prevent collisions in simulation. (2) vehicles are spaced out according to a certain spacing equation behind the leader in simulation[10]. The initial formula of the model is: 2(t) 2 (t) vn vn−1(t +b0 + + −T) = −T) ∆x(t βvn βl α (2) Where, ,b0 , , β1 β α are parameters, and the meanings of other symbols are the same as the former. The primary properties of the model are expressed as follows[5]~[11]: (1) Drivers’ “perception threshold” is not considered. (2)Length of the simulation clock steps can not be decreased easily in most cases, and the brake reaction time of drivers are restricted by it. (3) The model is appropriate for computer simulation. However, it can’t be used on intelligent vehicle to control vehicle’s operation in reality, because the discrete simulation clock step is adopted in the model, and the model updates the vehicles sequentially in the simulation, which conflicted with synchronous operation of vehicles in real world. (4) The hypothesis used to avoid collision is reasonable. To a certain extent, it can simulate car-following behavior in congestion state however, there are certain differences from the real vehicle operation. In reality, “retaining wall” brake is rarely used by the leading vehicle, because driver can make good use of the multi-resource information and react to the changes of the leading vehicle in time. In fact, drivers don’t keep the safety distance in most cases. Therefore, if traffic capacity analysis based on the safety distance car-following model is used, it will be very difficult to consistent with real maximum traffic volume. 2.3. AP model In 1963, some underlying psycho-physics driving factors were analyzed, and the action point model was