The Exploratory Advanced Research Program

Making Driving Simulators More Useful for Behavioral Research

Simulator Characteristics Comparison and Model-Based Transformation

BEHAVIORAL FIDELITY

Behavioral fidelity refers to the simulators’ ability to replicate driver behavior observed in the real world and is considered the ultimate measure of simulator fidelity. Researchers for this project collected simulator-based data and on-road data to describe behavioral fidelity for each of the four simulators used in the study. They collected data both with and without the simulators’ motion base engaged and with and without a more complex visual scene.

Data Collection

The analysis involved 167 participants ranging in age from 25 to 45 years. Forty-eight people each participated in the WTI, FHWA, and NADS simulators, and 23 people participated in the miniSim. The simulators operated on three different software platforms, but all used the same scenarios. Scenarios were used that involved two types of road segments, roundabout and gateway. The term gateway refers to a transition from a rural road into a town. In this study, the gateway was designed to achieve a 40-km/h (25-mi/h) speed limit on a two-lane suburban roadway in Iowa by using converging pavement markings, narrow lane markings, and speed advisories, as shown in figure 5. The roundabout scenarios were located on a rural two-lane arterial highway adjacent to the overpass of a major four-lane highway in Maryland, and a sequence of two roundabouts located on a rural two-lane roadway connected with a two-lane frontage road adjacent to an interstate highway in Arizona.^18,19

Figure 5. Example of geometric matching of real (top) and simulated (bottom) roadway geometry of a gateway in Iowa.

The research team selected these roadway elements based on discussions they had with FHWA about the potential application of driving simulators to investigate design issues.⁷ The researchers selected real-world examples of each road segment based on the availability of spot-speed data (i.e., the instantaneous speeds of vehicles at specific spots of the roadway) from published reports, and they based the virtual environment reproductions on the engineering schematics available for each site. The goal of these reproductions was to duplicate the road segment geometry and road features visible to the driver that were important to navigating the road segment; however, implementing identical scenarios on four different simulator platforms presented unique challenges and required a significant amount of fine tuning. A lack of established standards for road networks and dynamic element scripts was identified as a key impediment to sharing data on the driving environments across simulators as an efficient and smooth process.

The researchers used the same general procedure to govern data collection at each simulator facility, although minor variations were introduced depending on the logistical and operating requirements of each site. They used existing databases and local advertisements to recruit participants, who were initially contacted and verbally screened for eligibility and motion sickness before moving forward. Participants initially conducted a practice drive in the simulator to familiarize themselves with the operation of the simulator vehicle and experience of driving within the virtual environment, as shown in figure 6. Participants then drove the main route twice with varying levels of visual complexity and road segment orders.

Simulator Evaluation

Each participant then completed a simulator realism survey to evaluate the overall feel, braking response, and visual realism of the simulator. These components formed the three dimensions of subjective simulator fidelity, and the results showed that the simulators differ considerably over these three dimensions of simulator realism. The NADS simulator, with the advanced motion base on, had the highest reported realism of overall feel. The FHWA simulator represented the lowest realism of overall feel; however, this simulator did provide the highest perceived realism for being able to read the signs and see the road. This was attributed to the simulator being specifically developed to offer visual properties to support research for road and signage design. Drivers felt best able to brake and stop with NADS while having the motion on and with the WTI while having the motion off. This part of the study showed that no simulator configuration dominates the other in terms of perceived realism. In summary, the participants judged that the NADS is the most realistic simulator overall, the FHWA simulator best supports drivers’ ability to read signs, and the WTI simulator provides the best braking feel. Moreover, the effect of the motion base was shown to be relatively small and can even have a negative impact on realism.

Figure 6. An overview and screen image of the practice driving route.

Influencing Driver Behavior

As part of this research project, the research team analyzed the effects of simulator, motion, and visual complexity on driver behavior. The effect of motion failed to reach statistical significance and had little impact; however, visual complexity had a substantial influence on behavior. In some cases, particularly for drivers in the miniSim in the Iowa gateway, it led to an approximately 8-km/h (5-mi/h) speed reduction. In addition, drivers of the miniSim on the Maryland roundabout were traveling almost 16-km/h (10-mi/h) faster than were drivers in the other simulators. In most other cases, the influence of visual complexity was modest, so the substantial simulator differences between the miniSim and the other simulators is thought to have contributed to this speed difference.

In general, the collected data for the mean speed of drivers in the simulators relative to the mean speed observed on the road was similar. Drivers in the WTI simulator drove closest to real-world speeds at low speeds, whereas drivers in the FHWA simulator drove substantially slower at speeds below 50 km/h (31 mi/h) but drove faster once speeds increased beyond that threshold. Results indicated that speed varies more in the simulator than on the road. One explanation for this is that the simulator provides poorer cues regarding the speed and poorer feedback regarding drivers’ modulation of speed, leading to greater reliance on the speedometer, poorer speed control, and more variability in speed. This suggests that simulators can provide good estimates of the mean speed but poorer estimates of other elements of the speed distribution. Drivers drove faster and more variably in the miniSim and more slowly in the NADS relative to the speeds observed on the road, suggesting that the breadth of distribution may be more indicative of simulator fidelity than the mean speed.