TRAVERSE CITY, Mich. -- Self-driving cars still have a lot to learn before they can hit the road. And not all of it makes sense.
In a presentation at the 2017 Management Briefing Seminars on Wednesday, Melissa Cefkin, principal researcher of human-centered systems at Nissan's Research Center in Silicon Valley, said autonomous vehicles need to understand learned social behaviors — such as assuming a pedestrian standing outside the left side of the car is about to get in and drive and not walk into the street — before manufacturers can safely deploy them.
"We need to teach vehicles so they can move in a socially acceptable manner," Cefkin said in an interview.
While most human driving behaviors appear to be inherent, many — such as waving another car through an intersection — are actually learned through experience, she said. The challenge lies in programming these norms and behaviors into a self-driving car.
It is an addressable obstacle, Cefkin said.
"It's not the death of the situation but rather an opportunity to advance it," she said.
When autonomous vehicles start appearing on public roads, most people will experience them from the outside, either sharing the road as a driver of another car, or as a biker or pedestrian.
Nissan is addressing this education challenge by analyzing human driving behavior to determine which reactions are natural and which are learned. Once these behaviors are differentiated, developers can teach the vehicle what to expect in situations that don't adhere to natural assumptions, such as driving through a pedestrian-heavy college campus where people are less likely to abide by traffic signals.
"What we're trying to do is accelerate the ability to see those kinds of relationships that happen between different kinds of people in these situations," Cefkin said.
However, identifying such situations is incredibly complicated. In a four-way intersection alone, there are myriad right-of-way dilemmas that crop up, both between vehicles and between vehicles and pedestrians. How a car reacts to a situation also depends on time and place; traffic flows change by the hour, and each city has distinctive driving behaviors.
Cefkin predicts it will take a few years before driverless vehicles can adapt to learned driving behavior in an acceptable way, and even then it will be an ongoing process.
"We have years of development ahead to continue to work it out," she said.