COPENHAGEN, Denmark — Self-driving cars contribute to traffic congestion and could potentially be hazardous due to their inability to comprehend human behavior. A new study indicates that these autonomous vehicles, often promoted as the transportation of the future, struggle to interpret the subtle human social signals that inform driving decisions.

A key example of this issue focuses on deciding whether to yield or proceed in traffic – a decision that humans typically make swiftly and intuitively. However, self-driving cars consistently fail to interpret human behavior in traffic. As a result, their reactions can lead to traffic congestion and aggravate other road users, according to the award-winning study from a team at the University of Copenhagen.

“The ability to navigate in traffic is based on much more than traffic rules. Social interactions, including body language, play a major role when we signal each other in traffic. This is where the programming of self-driving cars still falls short,” says Professor Barry Brown, from Copenhagen’s Department of Computer Science, who has been studying the evolution of self-driving car behavior on the road for the past five years, in a university release. “That is why it is difficult for them to consistently understand when to stop and when someone is stopping for them, which can be both annoying and dangerous.”

Various companies, such as Waymo and Cruise, have initiated taxi services employing self-driving cars in certain regions of the United States. Tesla has deployed their Full Self-Driving (FSD) model to approximately 100,000 volunteer drivers in the U.S. and Canada. Despite these advancements, Prof. Brown and his team note that the actual road performance of these vehicles is a closely guarded trade secret with limited public insight.

Videos capture autonomous cars endangering pedestrians

To address this knowledge gap, the researchers conducted an in-depth analysis of 18 hours of 70 different YouTube videos, captured by enthusiasts testing cars from the back seat. One of the videos features a family of four waiting to cross a residential street in the United States. Despite the absence of a pedestrian crosswalk, the family wishes to cross the road. As the driverless car approaches, it slows down, prompting the two adults in the family to gesture for the car to continue. However, the car stops next to them for 11 seconds. As the family starts to cross the road, the car begins to move again, causing them to leap back onto the sidewalk.

“The situation is similar to the main problem we found in our analysis and demonstrates the inability of self-driving cars to understand social interactions in traffic. 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,” says Prof. Brown.

In tech-focused San Francisco, autonomous vehicles have been deployed as buses and taxis in various parts of the city, negotiating the hilly streets amid people and other natural elements. These vehicles cause traffic issues and disruptions in San Francisco due to their inappropriate reactions to other road users.

“I think that part of the answer is that we take the social element for granted. 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,” suggests Prof. Brown. “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.”

This study was recognized with the best paper award at the 2023 CHI Conference on Human Factors in Computing Systems and is published in the journal Association for Computing Machinery.

Could an extra traffic light help fix the problem?

Recently, researchers from North Carolina State University proposed adding a fourth color to traffic lights, a white light, enabling self-driving vehicles to help control traffic flow, and let human drivers in on what’s going on. Across a series of computational simulations, study authors found that adding this fourth white light significantly improved travel time through intersections and reduced fuel consumption.

“This concept we’re proposing for traffic intersections, which we call a ‘white phase,’ taps into the computing power of autonomous vehicles (AVs) themselves,” says Ali Hajbabaie, corresponding author of the paper and an associate professor of civil, construction and environmental engineering at NC State, in a university release. “The white phase concept also incorporates a new traffic signal, so that human drivers know what they are supposed to do. Red lights will still mean stop. Green lights will still mean go. And white lights will tell human drivers to simply follow the car in front of them.”

This “white phase” traffic intersection concept was actually first introduced by the team at NC state in 2020. However, that initial concept focused on using a centralized computing approach, with the computer controlling the traffic light totally responsible for receiving input from all approaching AVs, making the necessary calculations, and ultimately directing the AVs on how they should proceed through the intersection. That’s a lot of work for one computer.

“We’ve improved on that concept, and this paper outlines a white phase concept that relies on distributed computing – effectively using the computing resources of all the AVs to dictate traffic flow,” Prof. Hajbabaie comments. “This is both more efficient, and less likely to fall prey to communication failures. For example, if there’s an interruption or time lag in communication with the traffic light, the distributed computing approach would still be able to handle traffic flow smoothly.”

You might also be interested in:

South West News Service writer Jim Leffman contributed to this report.