Can AVs Interpret Human Behavior?
Oct. 27, 2020—When you’re behind the wheel and a pedestrian waves at you, what does that mean? Most would assume the pedestrian is giving you and your vehicle the right of way, signaling to you that should turn before they cross.
How does the interpretation change when it is a cop waving at you? Or a crossing guard? Or a construction worker? More often than not, humans understand other humans even when they are not communicating verbally, but can the same be said for a car?
Sam Anthony, co-founder and chief technology officer of Perceptive Automata says body language is indicative of intention. “The way someone is holding a bag is incredibly diagnostic as to whether or not they want to cross the road.”
Body language and facial expressions are highly communicative features for humans, but how does a driverless car interpret these everyday expressions? Partners for Automated Vehicle Education recently hosted a webinar discussing the integration of human behavior into autonomous vehicles. Getting one to inform the other is an uphill battle, but experts in the field gathered to share insights and challenges that they have identified so far.
Perceived Human Behavior
As a human driver, when you see someone standing at a crosswalk, you can generally figure out if they are planning to cross or not based on body language. Someone who is standing further from the intersection and looking at their phone is probably not about to dart into oncoming traffic—but how would a car know that?
Anthony says in the broader sense, his company is focused on human intuition for machines.
“What we try to do is use the techniques of behavioral science, then deploy a system in an autonomous vehicle that sees them in the same way.” Meaning a car could interpret an individual’s intention similar to how a human does.
“Any information that is available, people will use,” Anthony says. “The way someone is holding a bag is incredibly diagnostic as to whether or not they want to cross the road.”
Actual Human Behavior
When it comes to analyzing actual human behavior, Nadia Anderson, manager of government relations at Cruise, says a dynamic approach is the first step.
Pedestrian types range from the classic cyclists, joggers, and rollerbladers to today’s electric scooter rider. Each of these pedestrians will move, think, and react in various ways and autonomous vehicles should be prepared for any outcome.But Anthony says it is not as straightforward as it sounds.
“Pedestrians and cars have 100 years of history, they’re deeply learned,” he says. “Whereas an electric scooter, even the people riding them don’t know how they’re supposed to behave.”
At Cruise, Anderson says they are studying just that through the use of video games.
“How does Lebron move the same on the court as he does in a video game?” she poses.
It is through research of body language, movement styles, and above all, intention. Anderson says that is what they are studying at Cruise and will hopefully be able to teach autonomous vehicles.
“Humans are diverse and behave in diverse ways, in order to get this right, a diverse set of technologies and methodologies are needed,” Anderson says.
Infrastructure as a Role
Anderson says infrastructure absolutely plays a role in educating autonomous vehicles about human behavior. In fact, some streets in San Francisco have been designated as “slow spaces” where cars and other large vehicles are prohibited.
According to the San Francisco Municipal Transportation Agency, the goal of these spaces is “to make San Francisco more welcoming and accessible for people who want to travel on foot, bicycle, wheelchair, scooter, skateboard or other forms of micromobility.”
For shops this could mean a decrease in the frequency of collision repairs due to the extra safety measures and dedicated laneage to pedestrians. It could also mean an increase in the complexities of repairs should autonomous vehicles become commonplace on public roads.
The changing landscape of transportation is informed, namely, by technology.
“With technology, you can get more predictability,” says Anthony. When it comes to informing autonomous systems of human behavior, he says it’s more of an art than a science.