An errorable car-following driver model is presented in this paper. An errorable driver model is one that emulates human driver's functions and can generate both nominal (error-free) as well as devious (with error) behaviors. This model was developed for evaluation and design of active safety systems. The car-following data used for developing and validating the model was obtained from a large-scale naturalistic driving database. The stochastic car-following behavior was first analyzed and modeled as a random process. Three error-inducing behaviors were then introduced. First, human perceptual limitation was studied and implemented. Distraction due to non-driving tasks was then identified based on the statistic analysis of the driving data. Finally, time delay of human drivers was estimated through a recursive least square identification process. By including these three error-inducing behaviors, rear-end collisions with the lead vehicle could occur. The simulated crash rate was found to be similar but somewhat higher than that reported in traffic statistics.

Collision warning/collision avoidance (CW/CA) systems must be designed to work seamlessly with a human driver, providing warning or control actions when the driver's response (or lack of) is deemed inappropriate. The effectiveness of CW/CA systems working with a human driver needs to be evaluated thoroughly because of legal/liability and other (e.g. traffic flow) concerns. CW/CA systems tuned only under open-loop manoeuvres were frequently found to work unsatisfactorily with human-in-theloop. However, tuning CW/CA systems with human drivers co-existing is slow and non-repeatable. Driver models, if constructed and used properly, can capture human/control interactions and accelerate the CW/CA development process. Design and evaluation methods for CW/CA algorithms can be categorised into three approaches, scenario-based, performance-based and human-centred. The strength and weakness of these approaches were discussed in this paper and a humanised errable driver model was introduced to improve the developing process. The errable driver model used in this paper is a model that emulates human driver's functions and can generate both nominal (error-free) and devious (with error) behaviours. The car-following data used for developing and validating the model were obtained from a large-scale naturalistic driving database. Three error-inducing behaviours were introduced: human perceptual limitation, time delay and distraction. By including these error-inducing behaviours, rear-end collisions with a lead vehicle were found to occur at a probability similar to traffic accident statistics in the USA. This driver model is then used to evaluate the performance of several existing CW/CA algorithms. Finally, a new CW/CA algorithm was developed based on this errable driver model.

Advanced driver assistance systems, such as unintentional lane departure warning systems, have recently drawn much attention and efforts. In this study, we explored utilizing the nonlinear binary support vector machine (SVM) technique to predict unintentional lane departure, which is innovative as the SVM methodology has not previously been attempted for this purpose in the literature. Furthermore, we developed a two-stage training scheme to improve SVM's prediction performance in terms of minimization of the number of false positive prediction errors. Experiment data generated by VIRTTEX, a hydraulically powered 6-degrees-of-freedom moving base driving simulator at Ford Motor Company were used. All the vehicle variables were sampled at 50 Hz and there were 16 drowsy drivers (about three-hour driving per subject) and six control drivers (approximately 20 minutes driving each). A total of 3,508 unintentional lane departures occurred for the drowsy drivers and 23 for the control drivers. Our study involving these 22 drivers with a total of over 7.5 million prediction decisions demonstrates that: (1) Downloaded by [La Trobe University] at 03:56 16 June 2016 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 2 excellent SVM prediction performance, measured by numbers of false positives (i.e., falsely predicted lane departures) and false negatives (i.e., lane departures failed to be predicted), wereachieved when the prediction horizon was 0.6 s or less, (2) lateral position and lateral velocity worked the best as SVM input variables among the nine variable sets that we explored, and (3) the radial basis function performed the best as the SVM kernel function.

No abstract

an errorable car-following driver model was presented in this paper. This model was developed for evaluating and designing of active safety technology. Longitudinal driving was first characterized from a naturalistic driving database. The stochastic part of longitudinal driving behavior was then studied and modeled by a random process. The resulting stochastic car-following model can reproduce the normal driver behavior and occasional deviations without crash. To make this model errorable, three error-inducing behaviors were analyzed. Perceptual limitation was studied and implemented as a quantizer. Next, based on the statistic analysis of the experimental data, the distracted driving was identified and modeled by a stochastic process. Later on, time delay was estimated by recursive least square method and was modeled by a stochastic process as well. These two processes were introduced as random disturbance of the stochastic driver model. With certain combination of those three error-inducing behaviors, accident/incident could happen. Twenty-five crashes happened after eight million miles simulation (272/100M VMT). This simulation crash rate is higher by about twice with 2005 NHTSA data (120/100M VMT). ©2008 AACC