Driving Simulation for Evaluation of Driver Assistance Systems and Driving Management Systems

Development of advanced driver assistance systems and driving management systems requires careful and thorough evaluation of not only algorithms, but also user acceptance and adaptation. For that purpose, driving simulation provides an ideal environment by creating various driving situations that may not be possible in a test track for safety reasons and putting drivers in a simulation loop to evaluate objective performance and subjective feelings. This paper describes our on-going effort to expand our driving simulation capability to evaluate driver assistance and driving management systems. First, we present our driving simulator and its major upgrade to provide better driving simulation environment. Second, we report case studies on Adaptive Cruise Control (ACC) and a safe driving management system. We have conducted a series of simulator experiments on ACC to determine effects of both positive and negative behavioral adaptation by the drivers. We have found that ACC draws consistency in headway-time regardless of drivers’ driving styles. However, ACC also induces drivers’ blind reliance and distraction, resulting in reduced lane keeping ability, larger head and eye movement, and slower response to simulated ACC failure. We also report progress being made on implementing the safe driving management system on the simulator and evaluating the effectiveness of the dangerous driving detection algorithm in a variety of driving situations. DSC 2007 North America – Iowa City – September 2007 Introduction Advancement of vehicle electronic technology has led to development of various systems for assisting drivers and managing driving to improve vehicle safety and driver comfort. Driver assistance systems in general monitor driving situations continuously and take necessary actions to avoid possible accidents even without drivers’ intervention. Driving management systems store driving data and analyze driver behaviour and accidents to promote safe driving and identify accident causes. Development of those driver assistance systems and driving management systems requires careful and thorough evaluation of not only algorithms, but also user acceptance and adaptation. For that purpose, driving simulation provides an ideal environment by creating various driving situations that may not be possible in a test track for safety reasons and putting drivers in a simulation loop to evaluate objective performance and subjective feelings. This paper describes our on-going effort to expand our driving simulation capability to evaluate driver assistance and driving management systems. First, we present our driving simulator and its major upgrade to provide better driving simulation environment. We then describe case studies on Adaptive Cruise Control and a safe driving management system. Kookmin University Driving Simulator The driving simulator at Kookmin University has had substantial modifications since its first development in 1997, from a single seat simulator with a single channel visual system and a six DOF hydraulic motion platform to a half-car simulator with a three channel visual system and a fixed base in 1998 to a full-car simulator with a four channel visual system and a two DOF electric motion platform in 2001. Our driving simulator had another major upgrade this year (Figure 1). The motion system had the most substantial change. We replaced a traditional motion platform with four electric motors and links installed at individual corners of the car body. The new motion system simulates suspension movement realistically and accurately. The system is effective and fast in creating special effects such as bumps and rumble strips as well as roll, pitch and heave cues. We believe that motion envelop of a traditional motion platform must be substantially large in order to take full advantage of a motion washout algorithm. The washout algorithm can be more effective in generating correct cue and simulating sustained acceleration when implemented on a motion platform with sufficiently large strokes. We also installed new visual computers, DLP projectors and a display screen to improve performance of the visual system. We installed a new steering system that provides ______________________________________________________________________________________ DSC 2007 North America – Iowa City – September 2007 highly realistic steering feedback. We implemented Controller Area Network (CAN) for data transfer and reworked all wirings in the control force loading system. The major features of the new simulator are summarized as: (1) a four channel visual system that provides 140x40 and 50x40 degrees of front and rear fields of view, (2) a fast-response, 4-axis electric motion platform that produces roll, pitch and heave motion, (3) a full-car cabin with realistic control loading and full instrumentation, and (4) human performance measuring equipment including a head and eye tracking system and a physiological signal measuring device. Fig. 1 Kookmin University Driving Simulator Adaptive Cruise Control Study Adaptive Cruise Control (ACC) is a representative driver assistance system. ACC automatically adjusts vehicle speed, if necessary, to maintain a desired distance to a preceding vehicle, and thus improves driver comfort and safety. ACC, as an automated system, induces drivers’ trust and adaptation, which brings both positive and negative effects. ACC reduces driver error and accident possibility. ACC also reduces the number of sudden accelerations and decelerations, enables speed synchronization among vehicles, and encourages smooth lane change behaviours [1, 2]. On the other hand, drivers may use any freed visual, cognitive and physical resources to engage in non-driving tasks. These tasks may reduce their vigilance and attention to the primary driving task, which could result in driver distraction, and a failure to detect and respond to critical driving situations. ACC also deteriorates driving performance: Drivers’ lane keeping ability reduces and the drivers tend to brake harder and more often [3-5]. The objective of our simulator study is to develop a strategy to compensate negative behavioural adaptation to ACC. Possible approaches are to adjust assistance level depending on the extent of negative adaptation and to add lane keeping assistance capability. We first conducted a series of simulator experiments under normal driving and ACC failure situations to investigate driver behaviour [6, 7]. ______________________________________________________________________________________ DSC 2007 North America – Iowa City – September 2007 Normal Driving Experiment Forty drivers participated in the experiment to drive the simulator with and without ACC. When driving with ACC, they were instructed to use a cruise control button on a steering wheel to select the most comfortable headway-time between 0.5 and 2.5 seconds with an interval of 0.5 seconds. When driving without ACC, they were asked to follow a preceding car, while keeping small, but safe distance. Table 1 summarizes the experiment results for headway-time and standard deviation of lateral position of the car. The drivers maintained headway-time of 1.49 seconds when driving with ACC less than 2.31 seconds when driving without ACC. This implies that the drivers trusted ACC capability of controlling speed and distance, and felt comfortable with keeping shorter distance with ACC. In addition, the standard deviation of headwaytime is very small when driving with ACC. This shows that ACC draws consistency in driving speed and safe distance regardless of the drivers’ driving styles. The standard deviation of the lateral position of the car becomes larger when driving with ACC, although not significant. This is consistent with other researchers’ findings. This implies that the drivers’ lane keeping ability degraded, showing negative effect of behavioural adaptation to ACC. Table 1 Headway-time and Standard Deviation of Lateral Position for Normal Driving Experiment Driver performance ACC on ACC off M SD M SD Headway-time (sec) 1.49 0.05 2.31 0.48 Standard deviation of lateral position (m) 0.71 0.06 0.64 0.09 * M: Mean, SD: Standard deviation Figures 2 and 3 show head and eye movement areas of the drivers. The areas were obtained by projecting head and gaze direction vectors on the plane parallel to the driver’s face. When driving with ACC, both head and eye movement areas become larger and more dispersed. This shows the effect of driver distraction during driving. This is consistent with larger deviation of lateral position in Table 1.