Evaluation Method of ARAS Combining Simulator Experiment and Computer Simulation in Terms of Cost-Benefit Analysis

Advanced Driver Assistance Systems (ADAS) for four-wheeled vehicle safety such as Forward Vehicle Collision Warning System (FVCWS) and Lane Departure Warning System (LDWS) are already in practical use. Likewise, ADAS using Information and Communication Technology (ICT) such as Vehicle-to-Everything communication (V2X) are currently being developed [1‐3]. Owing to the spread of such active safety systems, collision accidents of four-wheeled vehicles are decreasing year by year [4, 5], yet collision accidents of two-wheeled vehicles are still occurring with high rates [6].

In the case of two-wheeled vehicles, Riding Simulators (RS) have also been developed to train novice riders and to analyze riding safety [7, 8]. Since 2008, Advanced Rider Assistance System (ARAS) and On-Bike Information Systems (OBIS) have been actively studied by the SAFERIDER project [9] for two-wheeled vehicle riders using RS. The SAFERIDER team studied five key ARAS functions: velocity alert, curve warning, frontal collision warning, intersection support, and lane change support. However, there is little previous research on ARAS for two-wheeled vehicles [10, 11], and it is not an active field compared with studies for four-wheeled vehicle safety. Thus, even though the vehicle dynamics and severity of injury in collision accidents are significantly different from four-wheeled vehicles, ARAS evaluation methods for two-wheeled vehicles have not been studied. Therefore, we are focusing on an evaluation method for upcoming ARAS for two-wheeled vehicles. This study focuses on the collision warning system of an ARAS for two-wheeled vehicles, and evaluates the accident-reduction effect through combing RS experiments [12, 13], computer simulation and cost-benefit analysis. The simulation was performed to quantify the accident-reduction effect using a model that can evaluate comprehensive safety as a human-machine system. The accident-reduction effects can be quantified by combining RS experiments and computer simulation. But quantification of collision-reduction effects is not sufficient to evaluate the societal benefits of introducing the system into the market for practical use. Although there are some studies that evaluated the accident-reduction effect of such systems [14], there are few evaluation methods that consider the social benefits. Cost-benefit analysis is typically used in evaluating the public utility of certain services or systems as a predictor which evaluates whether a new service or system will become beneficial to society, or not.

In this study, we quantitatively evaluate the effects of the ARAS not only in terms of accident-reduction effects of the system based on RS experiments and computer simulation methods, but also the social benefit of introducing ARAS to society. To evaluate the usefulness of the system according to the benefits of practical use in society, we propose an evaluation method for ARAS combining riding simulator experiments and computer simulation in terms of cost-benefit analysis.

This study has the following two purposes:

This purpose aims to quantify not only collision-reduction effect through the usage of ARAS but also to quantify the total reduction of accidents which combines collision due to both human errors and system errors. We developed an RS and investigated the collision-reduction effects of ARAS and riding behaviors (see Sect. 2). In addition, accuracy of quantifying the collision-reduction effects was improved by using a time-series error model (see Sect. 4.1). The total collision accident probability was calculated by quantified summation of collision probability of both human errors and system errors based on the integrated error model (see Sect. 4.2).

This purpose aims to quantify the social benefit generated by practical use of new ARAS for two-wheeled vehicles. For this purpose, we quantify the social benefit and cost. Benefit was calculated as the summation of the estimated value of the reduction of human loss and material loss. Human loss was calculated according to the severity of injury due to collision accidents based on simulated collision velocity (see Sect. 4.3). Material loss was calculated based on the average repair cost of vehicles according to each target collision accident pattern. Social cost is defined as the total cost of system introduction in this paper. Finally, we evaluate whether the system is beneficial to society by comparing the cost and benefit (see Sect. 5).

We investigated riding behaviors in terms of parameters that are related to the risk of traffic accidents including Reaction Time (RT), Average Deceleration (AD). In this study, RT means the time for the rider to start braking after detecting dangerous objects or risky situations. This is a very common and important parameter in the studies of traffic engineering and traffic accidents. Riders can prevent accidents by shortening their RT. AD indicates the average value of deceleration starting from responsive brake operation until the vehicle stops completely and it is also a critical parameter when braking operation in risky situations. If AD is high, the rider can stop in short time in a short distance. This parameter is highly relevant to the RT. If RT is short, the rider can quickly detect risky situations and stop the vehicle with safe deceleration.

Computer simulation of riding behaviors using experiment results data was carried out to compensate for a statistically insufficient amount of data from the riding simulator experiment. We also quantified the total collision probability, considering both the collision probability related to human error and the collision probability related to system error. Figure 5 is a flowchart showing the riding simulator experiment and the computer simulation.

This process consists of four steps. First, we investigated riding behaviors of test subjects in RS experiments. Second, we performed Monte-Carlo simulation based on a time series reliability model to quantify collision probability and collision velocity when riders are in a risky situation. Third, we estimated the total collision probability based on the integrated error model as a human-machine system. Finally, the fourth step estimated the injury level of the riders in case of a collision accident according to the total collision probability and the collision velocity.

Generally, cost-benefit analysis is used to evaluate large-scale matters such as public utilities and political decisions like environmental policy [20, 21]. Through this analysis method, we can confirm whether the analysis subject has social benefits or not, in terms of the benefit ratio (benefit / cost). Likewise, we can also evaluate the newly proposed ARAS from a social standpoint. Therefore, we propose combining methods to include riding simulator experiments, computer simulation methods, and cost-benefit analysis to evaluate the accident-reduction effect from a social perspective.

In this study, “cost” represents the assumed price of an ARAS multiplied by the number of two-wheeled vehicles in Japan. We considered that introducing the system would reduce the number of traffic accidents, therefore reducing both human loss and material loss. This total reduced loss (human loss + material loss) is regarded as the “benefit”. The cost-benefit analysis conducted in this study has four steps.

Figure 8 shows a flow chart of cost-benefit analysis that follows the computer simulation steps in this study (Fig. 5). First, we investigated accident data. Second, we calculated the benefit according to insurance data and severity of injury. Third, we calculated the cost of system introduction. Finally, the fourth step is comparing the cost and benefit.

Step 1: Investigation of accident data.

The target accident pattern of the system used in this study was collision accidents at intersections, so we conducted a survey focusing on accidents that occurred at intersections while two-wheeled vehicles were traveling on priority roads based on the statistical traffic accident database as reported by the Institute for Traffic Accident Research and Data Analysis (ITARDA) [22] in Japan.

Step 2: Calculate the benefit according to insurance data and severity of injury.

To quantify the benefit, we estimated the amount of human loss and material loss that will be reduced by introducing ARAS into the market. Human loss was calculated by estimating the rider’s injury severity level based on collision velocity [23]. Severity is usually divided into 4 stages from S0 to S3. S0 is defined as no injuries, but S0 is not applicable in accidents involving two-wheeled vehicles. This is because two-wheeled vehicle riders are most likely to suffer injuries when accidents occur. Next, S1 is defined as mild injury, S2 is moderate injury, and S3 is severe injury. To estimate the severity of injury, we calculated human damage due to accidents based on automobile insurance data [24]. We used this data to calculate the amount of insurance money paid per person for each level of severity. In the insurance data from the general insurance association of Japan, we could find the total insurance payment of each injury severity level (From Severity 1 to 3) at a year. However, the specific amount of insurance payment for each severity per person was not found. So, we simply divide the total insurance payment from the number of the accident at a year. The dividend amount was assumed as an average insurance payment to a person when the accident occurred for each severity level. Table 2 shows the relationship between collision velocity and injury severity level according to a previous study [23] and the human loss of each severity level was calculated as the average amount of insurance payment per person of the severity level for the year in the insurance data [24].

Step 3: Calculate the cost of system introduction.

In addition, it is necessary to consider not only human loss but also material loss. Material loss was calculated based on the average repair amount of various two-wheeled vehicles. In this paper, the average motorcycle repair cost is set as JPY 600,000.

Step 4: Compare the cost and benefit.

Next, by combining the material loss and human loss, we can calculate the total loss. Reduction of this loss through introducing ARAS is the benefit. In other words, to calculate the total benefit, we multiplied the number of accidents for each level of severity and the human loss for each level of severity, both with and without using the system. We also calculated the material loss by multiplying the average repair amount by the number of accidents under each condition, both with and without using the system. Material loss of each condition and human loss are added to calculate the total amount of damage under each condition. Finally, the difference between the total damage amount with and without using the system is the benefit of introducing ARAS. We also considered the cost required for system installation, defined as system cost for society.

In this paper, we assumed that ARAS will be installed in all two-wheeled vehicles in Japan within 10 years. We also assumed the market penetration rate, the discount rate, and system price as required parameters for cost calculation. The penetration rate was assumed to be 10% each year. This rate determines how much the system will spread within the market in one year. The discount rate is also an important factor, because this rate estimates future prices according to the current value. For example, if the discount rate is 4%, a value of $100 will be equal to the current value of $96 twelve months later. In the field of economics, it is commonly assumed that the annual discount rate is 4%, so we also set the discount rate at 4%. System price was based on the following formula (10), and the price includes both the system and necessary infrastructure. We used these parameters to calculate the annual cost required for installing the system, as shown below. Here, Ac is annual cost, Sp is the system price, Dr is the discount rate, and Ep is an extension period.

Finally, we compare cost and benefit to clarify whether the accident reduction effect due to system introduction is socially beneficial or not.

In this study, we proposed an evaluation method for ARAS which combines a riding simulator experiment, computer simulation, and cost-benefit analysis. Through our research, we reached the following conclusions:

We conducted an evaluation experiment for a collision warning ARAS to quantify accident reduction effects through Monte-Carlo simulation using a time-series reliability model. In this simulation, we calculated the reduction of collision probability and collision velocity to estimate the social benefit. We also evaluated the accident reduction effect as a human-machine system using our original integrated error model. Results confirmed that introducing the system into the market would have an accident reduction effect even considering possible system malfunction.

We calculated both the human loss and material loss according to accident statistics data to estimate the benefit due to the accident reduction effect. In addition, we calculated the long-term cost using parameters which set the cost for introducing the system into the market over 10 years. Results confirmed that the benefit ratio exceeds 1.0 after 4 years of practical use, meaning that societal benefit can be gained through ARAS introduction. Using these results, we have proposed a method to evaluate ARAS that considers accident reduction effects and social benefit combining riding simulator experiments, computer simulation, and cost-benefit analysis.

In this paper, we proposed a multi-step method to evaluate ARAS from a social standpoint. Our method can evaluate ARAS with appropriate time and cost by using both RS experiments and computer simulation.

In order to perform computer simulation to supplement the statistical reliability of a limited number of experiments, we pre-processed data describing the riding behaviors of test subjects using statistical techniques. Pre-processing was necessary because the riding behavior data obtained in the experiments for simulations was not quite sufficient for meaningful computer simulation. Overall trends of riding behaviors of experiment test subjects were used to verify the probability distribution with maximum similarity, and the number of data was increased after curve fitting with the appropriate probability distribution. The central limit theorem suggests that at least 50 samples are required for statistical reliability in order to confirm overall trends of riding behaviors, but our method is efficient enough compared to the time and cost used in actual driving experiments. We believe this will increase the reliability of computer simulation, and we think our evaluation method is adequate for evaluating and comparing new ARAS.

In our next study, we will investigate the relationship of bio-signals and driving performance to develop a new ARAS. Furthermore, future studies will be performed using this system for other patterns of traffic accidents.