Abstract— The purpose of this study is to design multiple- input multiple-output ANFIS (MANFIS) models to simulate and predict the future state of the overtaking maneuver in real traffic flow for four different time steps ahead. These models are designed to predict the behavior for 1, 2, 4 and 6 time steps ahead. Each time step is equal to 0.1 second. In these models, important factors such as distance, velocity, acceleration and the movement angle of the overtaking vehicle are considered. In these models, for all the variables, instantaneous values are used and none of them is considered constant. The presented models predict the future value of the acceleration and the movement angle of the overtaking vehicle. These models are designed based on the real traffic data and validated at the microscopic level. The results show very close agreement between field data and models outputs. The proposed models can be employed ITS applications and the like. I. INTRODUCTION Driver’s error contributes to over 75 percent of road crashes. A high percent of these crashes are related to overtaking maneuvers due to its complexity. Intelligent transport systems (ITS) are under active development worldwide as a means of reducing loss of life. Since overtaking maneuver is a complicated maneuver and so many factors affect it, the automation of this maneuver has been considered to be one of the toughest challenges in the development of autonomous vehicles. Driver behavior can be categorized into three main behaviors; car following [1, 2], lane changing [3, 4] and overtaking [5, 6]. Here, the concentration of this study is on the overtaking behavior as one of the most challenging behaviors on highways. An overtaking maneuver consists of three phases: a) diverting from the original lane, b) driving straight in the adjacent lane, and c) returning to the lane. The phases of the overtaking maneuver are shown in Fig. 1. These three phases can be called in short: lane changing, overtaking and returning. From this point, it is indicated that the relation between lane changing and overtaking is intimate, lane changing is an important part of overtaking process, and it is the base of overtaking, because it is necessary to change lane before overtaking [7]. Ali Ghaffari is with the Mechanical Engineering Department, K. N. Toosi University of Technology, Tehran, Iran (e-mail: ghaffari@kntu.ac.ir). Alireza Khodayari is with the Mechanical Engineering Department, K. N. Toosi University of Technology, Tehran, Iran, (corresponding author, e- mail: arkhodayari@ieee.org). Fatemeh Alimardani is with the Mechatronics Engineering Department, K. N. Toosi University of Technology, Tehran, Iran (e-mail: f.alimardani@ee.kntu.ac.ir). Hossein Sadati is with the Mechanical Engineering Department, K. N. Toosi University of Technology, Tehran, Iran (e-mail: sadati@kntu.ac.ir). Fig. 1. Two lane changes during the overtaking maneuver. In this paper, multiple-input multiple-output ANFIS (MANFIS) models for modeling and prediction of the Driver-Vehicle-Unit (DVU) behavior in overtaking scenarios for different time steps ahead is presented. These models predict the future value of the acceleration and the movement angle of the overtaking vehicle. In real driving situations, driver can only control the steering wheel and the pedals. By turning the steering wheel, the movement angle of the vehicle varies, and by pushing one of the pedals, throttle or brake pedals, the acceleration of the vehicle changes. The remaining parts of this paper are organized as follows: Section II describes a brief review of the previously presented models. Section III presents the new overtaking models design. Four MANFIS models are designed based on real traffic datasets to predict the acceleration and the movement angle of the overtaking vehicle. These models predict the acceleration and the movement angle of the overtaking vehicle of 1, 2, 4 and 6 time steps ahead. Each time step is equal to 0.1 second since the data sampling time is equal to 0.1 second. In other words, these models can predict the acceleration and the movement angle of the overtaking vehicle of 0.1, 0.2, 0.4 and 0.6 seconds ahead. In Sections IV, the proposed models are evaluated and the conclusion is given in Section V. II. RELATED WORKS ON OVERTAKING MANEUVER MODELING The study of overtaking models has not been very extensive. In this section, a brief review on the few previously presented overtaking models is presented. In 2000, Polus et al. developed a model to estimate passing sight distance of the overtaking process [8]. In 2003, Naranjo et al. offered a rule which its inputs were the velocity of the two involved vehicles and its output was the overtaking distance [9]. In 2004, Shamir designed a smooth and ergonomic optimal trajectory for the overtaking maneuver [10]. In 2005, Hassan developed a mathematical model based MANFIS-Based Overtaking Maneuver Modeling and Prediction of a Driver-Vehicle-Unit in Real Traffic Flow A. Ghaffari, A. Khodayari, Member, IEEE, and F. Alimardani, H. Sadati 2012 IEEE International Conference on Vehicular Electronics and Safety July 24-27, 2012. Istanbul, Turkey 978-1-4673-0991-2/12/$31.00 ©2012 IEEE 387