EXPERTS ADVISORY

A driver of Tesla’s Model S premium electric vehicle equipped with an autopilot system was killed in a crash with a tractor-trailer that crossed in front of the car. The Tesla’s brake was not applied, since the automated system nor the driver recognized the white side of the semi-truck against a brightly lit sky.

Experts at the University of Michigan are available to discuss.

Anuj Pradhan is an assistant research scientist in the Human Factors Group at the U-M Transportation Research Institute. He is interested in the etiology of injuries and fatalities due to motor vehicle crashes from a human factors and behavioral standpoint, and has focused on the human factors issues associated with automated vehicles.

“Autopilot mode on Teslas is predicated on the driver always being ready to take back control from the system with essentially zero notice,” he said. “However, when drivers no longer actively drive the car (i.e., undertake a control task) but still have to read the road environment as if one were driving so as to take back control at any time (i.e., perform a monitoring task), their performance in the monitoring task rapidly deteriorates.

“Sustained attention and continuous monitoring without active participation has been long recognized as an extremely challenging task for a human to perform. In this tragic case, the failure of the sensors to detect the other vehicle was compounded by the driver himself apparently failing to detect the imminent crash and react accordingly.

“Since he was not actively participating in the driving task, the driver probably had reduced awareness of the driving situation. This is a recognized human factors issue in the domain of human-automation interaction, and significant research is being conducted worldwide to study how the driver can be monitored or kept reliably engaged so as to remain within the ‘driving loop.”

Contact: 734-647-9191, [email protected]

Jason Corso is an associate professor of electrical and computer engineering at U-M’s College of Engineering. He studies computer vision, including how computers understand scenes, as well as general image and video understanding.

“The computer vision technology behind autonomous and advanced driver assistance systems is a topic of intense study—even the Tesla autopilot is explicitly denoted as a Beta product,” he said. “While tragic, I would not consider this a major setback, so much as a wake-up call that significant further study is needed to model the sensors and the underlying recognition technologies on which these systems rely. This is an unfortunate reminder that we have more work to do.”

Contact: [email protected]

James Sayer is the director of the U-M Transportation Research Institute and serves as the project manager of the Connected Vehicle Safety Pilot Model Deployment, a U.S. Department of Transportation-sponsored program to demonstrate connected-vehicle technologies in a real-world, multimodal environment.

He can address the importance of vehicle-to-vehicle and vehicle-to-infrastructure technology and the importance of testing new technologies in dangerous scenarios in controlled environments like Mcity, U-M’s unique test facility for evaluating the capabilities of connected and automated vehicles and systems.

“Humans are really not particularly good at long-term monitoring, of doing a task that just requires them to stare at dials or gages for example,” he said. “So when you provide a level of automation that controls a vehicle the vast majority of time, they may get to the point where they are no longer monitoring as well as they should because they’re gaining all of this experience when the system is working properly.”

Contact: 734-764-4159, [email protected] (before July 5)

Michael Flannagan is a research associate professor in the Human Factors Group at the U-M Transportation Research Institute. His research expertise focuses primarily on driver vision and psychophysics.

“The (Tesla) crash scenario that has been described is a well-known perceptual problem for human drivers—in other words, seeing ‘the broad side of a truck’ is sometimes harder than one might expect—and as such should have been foreseen,” he said.

Contact: 734-936-1091, [email protected]