Road Friction Coefficient Estimation For Vehicle Path Prediction

One of the most noticeable trends of the automotive industry in the 1990's is the emerging of active safety technologies. Instead of passive protection and crash worthiness, emphasis has been shifted toward accident prevention and damage reduction using active safety techniques. To develop effective active safety devices, it is necessary to estimate vehicle motions accurately, and reliable road friction estimation is one of the most important steps toward this goal. This paper is part of an on-going research toward the development of a lane-departure warning system [1], one of the most technology-demanding active safety systems. The basic concept of the lane-departure warning system is to project vehicle trajectory, and compare with the perceived road geometry (obtained from vision systems) to calculate a performance metric termed "time to lane crossing" (TLC). When the calculated TLC is less than a threshold value, the control system will either issue warning signals or take intervention actions. To project the vehicle future trajectory accurately, we need to update vehicle parameters (speed, cornering stiffness, etc.) and estimate external disturbances (road super-elevation, wind gust, etc.). The estimation of external disturbances further depends on the vehicle parameters. Therefore, on-line estimation of road/tire characteristics is crucial for the TLC calculation and the overall lane-departure warning system. It is important to note that accurate road/tire friction estimation can also improve the performance of many other vehicle control/safety systems such as ABS, traction control, and 4WS systems. Recently, various methods to identify the road friction coefficients have been developed. Dieckmann [2] developed a method which allows the exact measurement of wheel-slip in the order of 10 . From the information of measured wheel slip the road surface variation is detected. Eichhorn and Roth [3] used optical and noise sensors at the front-end of the tire and stress and strain sensors inside the tire’s tread and studied both "parameter-based" and "effect-based" road friction estimation methods. Ito et al. [4] uses the applied traction force and the resulting wheel speed difference between driven and non-driven wheels to estimate the road surface condition. Pal et al. [5] applied the neural-network based identification technique to predict the road frictional coefficient based on steady-state vehicle response signals. Pasterkamp and Pacejka [6] developed an on-line estimation method based on lateral force and self-aligning torque measurements. In this paper, we have developed a disturbance observer to identify the road surface friction coefficient. The effect of the road surface condition on vehicle path prediction (measured in TLC) is the main performance evaluation metric. The vehicle path prediction is obtained based on the 2-DOF lateral dynamic model (bicycle model). To calculate the TLC, the longitudinal tire force is first estimated