Naohiko Tsuru

Multi-scale driving behavior modeling using hierarchical PWARX model

This paper presents a hierarchical modeling strategy of the driving behavior based on the multiple levels of abstraction of the underlying dynamics. In order to realize this idea, a new mathematical model, Hierarchical PieceWise ARX model (Hi-PWARX model) is proposed. The identification scheme for the Hi-PWARX model is developed by synthesizing the ideas of a feature vector definition revealing the dynamical characteristics and an unsupervised clustering technique. By applying the Hi-PWARX model to the observed driving data, a consistent hierarchical behavior model is obtained without any prior knowledge. The usefulness of the proposed modeling is demonstrated by the analysis of the driving behavior on the express way.

On driver's braking behavior in car following

A new index have been proposed to evaluate danger situation in car following based on the hypothesis that drivers evaluate collision risk for preceding car and decide braking behavior with change of area of a preceding car on the retina. Based on the index, braking behaviors in car following situation is analyzed. Decelerating profile of a test driver is modeled using the proposed index and the gap D based on the experimental results of the test driver with a real car. In addition, the results of laboratory experiments using a driving simulator support the results of the real car experiments. In addition, braking initiation timing of the test driver is characterized as a brake judgment line in the plane of the modified index and gap D and the characterization is applied to analyze change of the braking initiation timing according to the change of road condition.

A unified assisting system for longitudinal driving behavior based on model predictive control

Recently, there has been a growing demand for the design principle of assisting system. This paper presents a new framework of the assisting system design based on a continuous/discrete hybrid system model of the driving behavior and the model predictive control scheme. First of all, the driving behavior is identified as the piecewise ARX model, then it is explicitly embedded in the optimization problem as one of the constraints. Since the driving behavior includes some logical variables, the optimization problem is formulated as the mixed integer programming. The model predictive assisting control is realized by solving the optimization problem in real time. In addition, some adaptation mechanism to accommodate the change of the situation is discussed and verified.

Formulation of deceleration behavior of an expert driver for automatic braking system

In order to develop more effective automatic braking system or other braking support system, comfort of drivers should be taken into account. In this paper, deceleration behaviors of an expert driver in car following will be formulated as an example of comfortable braking behavior. The difficulty of its formulation is how to extract characteristics of the deceleration behavior. Expert driverpsilas braking behavior will be formulated using our proposed indices representing driverpsilas perceptual risk of collision from visual input. It will be shown that smooth deceleration profiles will be uniformly generated with very simple calculation for wide range of approaching conditions. Brake initiation timing will be also modeled by brake judgment line based on the proposed indices.

Steering-assist control system on curved road using car-to-car communication

Several driver-assist systems have been developed to enhance driver comfort, such as Adaptive Cruise Control (ACC) and Lane-Keeping Assist System (LKAS). Drivers must steer the vehicle even with the current ACC. LKAS manages the vehicle position within the lane in when the road curvature is not very large. The ability of such systems strongly depends on the recognition technology applied to the road environment. A sensor fusion method is one option to increase robustness of environmental recognition. The present study seeks to develop a road-information acquisition method based on a car-mounted sensor and a car-to-car communication system and to explore its application to an automatic steering-control system. In this paper, we assume that there is a lead vehicle in front of the host vehicle within the detection range of the car-mounted sensor and it can detect its relative position. First, a virtual road boundary estimation method is proposed based on measurement of the relative position of the lead vehicle and the lead vehicle status obtained by the car-to-car communication method. A steering-assist control method is then proposed using the estimated virtual road boundary information. Finally, the effectiveness of the proposed steering-assist control method is demonstrated by simulation results.

Formulation of braking behaviors of expert driver toward automatic braking system

Deceleration patterns of an expert driver will be formulated as an example of comfortable braking pattern. The difficulty of its formulation is how to extract characteristics of the deceleration behavior. This research focuses on visual control of vehicles in car following. We have hypothesized that driver evaluate risk of collision based on the area change of the preceding car on the retina and determines his/her braking operation. Based on the hypothesis, an expert drivers deceleration patterns will be formulated using the proposed risk perception index with visual information. It will be shown that the formulated braking pattern can generate smooth deceleration profile for many conditions of relative velocity and brake initiation timing uniformly with very simple calculation and without collision under some assumption. In addition, the model can determine brake initiation timing by specifying limitation of deceleration.