The Exploratory Advanced Research Program

Making Driving Simulators More Useful for Behavioral Research

Simulator Characteristics Comparison and Model-Based Transformation

INTRODUCTION

This report summarizes the results of a Federal Highway Administration (FHWA) Exploratory Advanced Research (EAR) Program-funded research project that explored the challenges of using driving simulators to guide roadway designs and evaluate design choices. The aim of this project was to help engineers identify the appropriate simulator platform for particular design questions, as well as to identify a mathematical transformation that can equate simulator data to real-world outcomes.

Highway and traffic engineers face considerable challenges in creating designs that are consistent with drivers’ capabilities, expectations, and limits.¹ Drivers often behave in complex and counterintuitive ways, and failing to consider driver behavior can cost lives and millions of dollars if roadways require revision after they are built.^2,3 Driving simulators provide a promising approach to addressing this challenge because they make it possible to visualize new roadway designs as well as safely expose drivers to demanding situations without the expense of fully implementing the design.⁴ Driving simulators also provide a means of conveying road design concepts to stakeholders through visualization and have the potential to be an important part of policy decisions and public acceptance.^5,6

Improving Understanding

There have been many recent advances in simulation technology, which has led to a wide range of driving simulators available to researchers. These simulators all offer different levels of realism, known as fidelity, in addition to varying levels of complexity and usage costs. Such diversity makes it difficult for researchers to know which simulator is appropriate to address a given design question. This uncertainty is thought to be one reason why simulators have not been more widely used by highway and traffic engineers.⁷

An improved understanding of the varying characteristics of simulators and how well they might reproduce driver behavior would make driving simulators far more useful for engineers. The ideal situation would be for simulator characteristics to exactly match actual cars and roadways, but this is beyond the capabilities of even the most advanced simulators at this time. Instead, the goal is to minimize the differences between the physical characteristics of the simulator and the roadway and therefore ultimately minimize the difference between behavior observed in the simulator and out on the road.

Understanding Physical and Behavioral Fidelity

In addition to physical differences, there are several other factors known to affect driver behavior that can prove difficult or impossible to simulate, including a driver’s motivation for the trip or the real-world consequences of a crash. Matching the physical features of the simulator to the roadway experience, known as physical fidelity, is therefore just one condition that must be replicated to ensure that driver behavior in the simulator matches behavior observed on the road.⁸ Until now, the driving simulator community has mostly focused on gross measures of physical fidelity, such as “high” or “low” fidelity. The next step, and broader research goal, is to match the behavior of drivers in the simulator with behavior on the road, known as behavioral fidelity.⁴ This requires sufficient realism of simulator controls and vehicle-handling characteristics to match actual vehicle performance.⁹ The goal is for behavior in the simulator to match behavior on the road accurately enough to support design decisions.

Simulator fidelity is further complicated by the fact that physical and behavioral fidelity are related to each other, in that imperfect physical fidelity will lead to imperfect behavioral fidelity.⁸ Despite this, imperfect fidelity is still often sufficient to support roadway design decisions. For example, a simulator might fail to accurately replicate the cues required to guide behavior, possibly leading drivers to drive faster than they would on the actual road.^10,11 Drivers, however, rely on multiple interchangeable cues to guide behavior and can substitute one set of cues for another.^12,13 This means that two different simulators might still produce similar behavior results because drivers can adapt and use the available cues within each simulator.^14,15 Simulator fidelity is further influenced by the level of information a simulator might provide for one task compared with another. A simulator might offer a high level of realism, or fidelity, for one set of tasks but only a medium level of realism for another. For example, a simulator may offer highly accurate renderings of road signs for a sign-reading task yet would be classed as a low-fidelity simulator for driving that involves 90-degree turns because it fails to provide a preview of the road on which drivers rely during the turns.⁴

Comparing Simulators and Scenarios

During this project, the research team explored task-dependent fidelity and examined the difference between physical and behavioral fidelity. The following summary compares behavior across four simulator platforms with four different configurations of motion base and visual complexity. The simulators were used to analyze a total of six roadway scenarios among them, comprised of four roundabouts and two gateways. This summary includes a description of the different driving simulators used in the project, a description of physical fidelity in terms of the cues drivers use for vehicle control, an assessment of the behavioral fidelity of these simulators, and an overview of a model developed as part of the project to relate simulator behavioral data collected in a driving simulator to data collected on the road.