Deciphering the more subtle choreography between pedestrians and drivers may help stem traffic deaths and provide guidance for engineers preparing self-driving vehicles.
The development of autonomous vehicles has been Domeyer's focus in research conducted first at the Massachusetts Institute of Technology and now at the University of Wisconsin, where he is studying how drivers and pedestrians signal their plans to each other.
His findings may enable engineers to someday address one of the more vexing aspects of autonomy — helping computer-controlled vehicles understand the intent of pedestrians and communicate their own planned actions.
Should those communications rely on traditional means, such as hand gestures and eye contact, then engineers must first understand how those work. While there's a wide belief those interactions depend on eye contact, that's not what Domeyer and his colleagues have found.
"Pedestrians actually look away about one second before they cross," he said. "There's a thing where they essentially say, 'I believe you, and I know you are going to stop, and I'll now look over here.' Which is unintuitive. In this case, it's the opposite of eye contact that's really important."
While pedestrians believe they make eye contact with the driver of an approaching vehicle, more than 90 percent cannot see a driver at all from 30 meters away — or determine their gaze from 15 meters away — according to an April 2019 report from Domeyer and other researchers at MIT's AgeLab.
That means a pedestrian's decision to cross is based more on perceived speed or deceleration of the vehicle.
In an automated-vehicle era, automakers and tech companies have experimented with ways to communicate an AV's intent to other road users. The now-defunct Drive.Ai installed external screens on its vans that would display messages such as "Waiting for you to cross." In February 2019, Ford Motor Co. tested how pedestrians responded to a roof-mounted light bar that projected visual cues.
But overall, figuring out how AVs should best communicate with others remains a fledgling endeavor.
At Wisconsin, Domeyer has analyzed pedestrian-vehicle interactions from a car's perspective. He's examining videos and Controller Area Network bus data to measure vehicle deceleration profiles and how they influence pedestrian activity.
Once he understands what constitutes "good" communication, he hopes his research can someday help software engineers write better control algorithms to allow self-driving cars to physically manifest proper communication. Overall, that underscores the complexity of these relationships.
"It's not getting the thing to navigate around an object," he said. "It involves people, and thus, there's a social component, and you can't dismiss things like politeness and fairness when we are talking about technologies that affect people's lives."