Researchers wanting to improve traffic safety often turn to various interventions based on education, traffic engineering, or intelligent transportation systems (ITS). One example might be a vehicle seat that vibrates on the left or right side whenever a driver strays out of the lane.
How successful these interventions are depends on how they perform in the actual traffic environment for which they were designed. However, testing that begins in the controlled conditions of laboratories and driving simulators often cannot be reliably replicated in a more naturalistic test-track scenario. That may change as a result of work by researchers with the ITS Institute's HumanFIRST Program. On a recent project with automaker Nissan, they developed a new method of reproducing conditions previously generated in a driving simulator in more natural, "real-world" driving environments.
Although traditional experimental environments are removed from real-world driving conditions, there are both scientific and practical reasons for developing interventions in a sequential research program that spans a range of research and testing environments, says Nicholas Ward, HumanFIRST Program director.
"From a scientific perspective, this research process benefits from the control you have in a laboratory or simulator to test broader theories before validating those conclusions on a test track or in the field," Ward says.
A critical requirement for transitioning research from controlled experiments in a simulator to more naturalistic studies in the driving environment is the ability to create comparable test scenarios for each of the different test environments. The test scenario defines the type of event that a participant experiences and what performance measures are used for evaluation.
Driving simulators offer a number of benefits, Ward says. They allow precise control over all elements of the driving environment to create identical and repeatable experiences. They also allow researchers to analyze risky scenarios without endangering a participant.
As a result, simulators are often used to design and evaluate safety interventions by developing crash scenarios that would not be feasible or ethical in the real world of test track and field studies. This means that the same scenarios generally are not used in the simulator and test track experiments in a research program.
In support of a progressive research program, the HumanFIRST Program developed a new method to transfer controlled simulator scenarios to test track (and field) studies with instrumented vehicles. One component of the method is an instrumented lead vehicle fitted with DGPS that operates within a digital map generated for the test environment. This vehicle is linked by wireless communication and DGPS to a test vehicle. Onboard computers produce specific driving scenarios, such as a specified speed profile with a determined braking event triggered either by a defined position on the test route or a set of conditions detected from monitoring the following test vehicle. These scenarios can be the same as those used in the driving simulator, as both are programmed with the same control logic.
In this application, a prior simulator study examined driver response in a rear-end crash scenario during which the driver of the following car was distracted with a secondary task. This involved linking an eye tracker and the secondary task to the simulator so that the car-following scenario could be automatically triggered when distraction was detected. The lead vehicle would follow a specified speed profile and then "hunt" for the subject in the following car to move within a defined range of headway (to create a desired hazard level). When the eye tracker detected that the driver was distracted with the secondary task, a trigger would result in braking with a specified deceleration rate.
In the test track study, the test vehicle was also equipped with an eye tracker and the secondary task. The gaze direction of the driver and interaction with the secondary task was again used to define distraction. The onboard computer, linked to the control actuators of the lead vehicle, was programmed to autonomously create a similar speed profile as used in the simulator study. The driver of the lead vehicle needed only to steer, since the parameters of all events were controlled precisely by the computer control system.
The lead vehicle also communicated with the test vehicle. At a determined point in the scenario, the lead vehicle would hunt for the test vehicle and also monitor the output from the eye tracker and secondary task. As soon as the driver was distracted, the lead vehicle executed its specified scenario-braking event. By virtue of the DGPS and digital map, the test vehicle recorded the same types of data as the simulator with comparable levels of reliability.
The second component of this method is a virtual "safety net." Rather than using a mechanical safety barrier to prevent a crash, the researchers devised a virtual barrier involving continuous communication between the two vehicles. An algorithm was developed to automatically trigger evasive acceleration in the lead vehicle and braking in the test vehicle if the headway was detected to move into a specified hazard region, Ward explains. In addition, a "watch dog" system continuously monitored the safety-critical components of the integrated system, and, if it detected failure, the system alerted the drivers with flashing lights and stopped both vehicles.
The combination of the automated lead vehicle and safety net provides an ecological way to translate scenarios from a driving simulator to real-world conditions with the same level of data reliability and control without risking safety, Ward says. Conversely, this system allows the researchers to replicate test track infrastructure perfectly within the simulated environment. Moreover, the use of the safety net extends the range of risk that can be developed in the test scenarios. "In particular, because subjects are not aware of the safety net, we can 'safely' observe driver responses to situations that they perceive to be risky," he says.
In these examples, the system is applied to a scenario with a pair of vehicles. However, Wards says the same protocols can be applied to supporting multiple-vehicle scenarios to allow controlled experiments of more complex traffic situations.