Navigating traffic can be complicated. We’ve all experienced stand-offs between pedestrians and cars, polite roundabout still-lives, and lengthy stares at drivers blocking an exit while queuing. It’s usually fine. People routinely solve ambiguous situations with eye contact and hand gestures – of one kind or another. But these social interactions elude self-driving cars.
More autonomous vehicles are joining the roads. In December 2020, Google’s self-driving car Waymo became the first to offer fully driverless taxi rides in San Francisco. Cruise, a General Motors subsidiary, launched a competing service in the same city in February 2022. Tech-enthusiastic car buyers are also served. In 2021, Honda sold the first level 3 ‘eyes off’ car to Japanese consumers, and Mercedes gained approval for sales in Nevada, USA. Today, these are still the only models at this level of autonomy, while Tesla remains at the lower level 2 ‘hands-off’ rating.
Progress is slow because self-driving cars don’t yet understand social interactions in traffic. Only six months ago, Ford and Volkswagen even scaled back their plans for autonomous vehicles. Barry Brown, Professor at the University of Copenhagen’s Human-Centred Computing department, explains the problem: “The driverless vehicle stops so as to not hit pedestrians, but ends up driving into them anyway because it doesn’t understand the signals. Besides creating confusion and wasted time in traffic, it can also be downright dangerous.”
Humans navigate the social element naturally and we ease. “We don’t think about it when we get into a car and drive – we just do it automatically. But when it comes to designing systems, you need to describe everything we take for granted and incorporate it into the design,” he elaborates.
Brown, along with Mathias Broth, Professor at Linköping University in Sweden, and Erik Vinkhuyzen of the Work, Interaction and Technology Group at King’s College, London, focused on the most basic and critical question when operating a vehicle: should I go or give way?
The team used video footage of Waymo rides and Tesla drives to understand how autonomous cars stop and interact with other road users. The researchers scrutinised more than 18 hours of mostly unedited recordings from 70 YouTube videos shared by the public. Interactions with pedestrians, it turned out, are the most awkward.
In one video, a family waiting on the pavement gesture for a Waymo to go ahead and turn in front of them. But the car stops and stays put. They stare at it for 11 seconds before the mother starts to cross after all. At that precise moment, the Waymo also moves. She scurries back to the pavement. The Waymo stops again, and a passenger shouts from the backseat, “Sorry, self-driving car!”
![A cartoon made of video stills where a Waymo car and passenger interact with pedestrians. In our gures we have used a comic book visualisation for transcribing the unfolding activity [42]. Video from JJ Ricks YouTube channel, 22 October 2020. Sorry, self-driving car!](https://i0.wp.com/www.bellwether.works/wp-content/uploads/2023/06/fig_1_pedestrian_family.png?resize=1024%2C727&ssl=1)
The pedestrians wait it out, and everyone is on their way after the 30-second ordeal ends safely. The Waymo system struggled, but it’s not the only one. In a similar incident, a Tesla froze the scene when someone hesitated to cross the road. The hands-off driver, however, could solve the problem with a simple wave.
Another recording shows a Tesla at an intersection. It’s their turn to go, but the car doesn’t move. The situation is complicated: the vehicle needs to move forward and wait for other cars to clear the junction before proceeding. But the software can only make simple decisions to go or not. The two-step problem is too difficult, and five cars pass before the Tesla inserts itself ungracefully.
The wrong decision may lead to a minor fender bender in this last scenario. But vulnerable road users like pedestrians and cyclists risk serious injuries. They rely on social interaction and traffic civility rather than just following the rules. The researchers conclude, “As we argue here, the challenge then is not one of computability but of understanding social interaction.”
“The car industry could learn from having a more sociological approach. Understanding social interactions that are part of traffic should be used to design self-driving cars’ interactions with other road users, similar to how research has helped improve the usability of mobile phones and technology more broadly”, Brown adds in an interview.
Source: Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems via EurekAlert!
Click here for more new developments in AI.