A driverless car mistook an advertisement on the side of a bus for a group of real people and performed an emergency stop, it has been revealed.
The poster – which was promoting the 2015 spy film The Man from U.N.C.L.E. – featured several actors who were identified by the vehicle’s AI system as pedestrians in the road.
John McDermid, a professor of software engineering at the University of York, revealed details of the incident during a briefing in London to discuss the technology.
This week, Waymo, the US driverless car firm, announced it would launch a pilot of its robotaxi service in London this year, while Wayve is expected to roll out the service with Uber later in the year.
However, British experts cautioned that the technology was not perfect. They said that during trials, driverless cars struggled to distinguish between real people and images, understand roundabouts, identify rule-breaking by pedestrians, and know when to slow down on a tight curve.
During the briefing at the Science Media Centre, McDermid said: “One of the automated vehicle companies I work with had a situation where their vehicle did a sudden emergency stop because it saw pedestrians in the road, except they weren’t.
“It was a life-size advert on the side of a bus, but to an AI, it was human beings. That seems very obvious [to us], but actually, to the AI, it’s not.”
McDermid also showed a video of automated car trials in York. He said the vehicles struggled when people made unexpected movements, such as crossing a road while the pedestrian light was red.
He said: “It’s seen that there’s a traffic light, so identified ... the hazard, because the light is red. It changes to green, the vehicle is about to move off. But this is York, so the tourists, although the lights change to green, they still walk across the road.
“Computer vision doesn’t understand what it doesn’t have models for in the world; it doesn’t know what a roundabout is.”
However, experts said that automation offered an opportunity to address some of the factors that contribute to road casualties.
More than 1600 people die on British roads annually, with road crashes one of the most common causes of death among young people.
Under the Automated Vehicles Act, passed in 2024, self-driving vehicles on British roads must achieve a level of safety equivalent to, or higher than, that of careful and competent human drivers.
But researchers said that while humans were very good at recognising the sanctity of life, and largely drove carefully to avoid harm to themselves and others, computer systems did not have such a moral code.
Dr Nick Reed, an independent expert consultant at Reed Mobility, said there was something “fundamentally different” between how AI systems viewed people and how humans did.
He said: “[Humans] understand that there are personal consequences if they get caught breaking the rules.
“These behaviours are inherently human and are based on millions of years of evolution. We can’t have the same confidence over the behaviour of automated vehicles.
“I think there is something fundamentally different between a machine labelling an object as [a] person and us as humans, recognising an individual as being a person, someone who can get hurt, someone who has family, someone who has hopes and dreams.
“We need to be sure that the machines with which we share public space, and the companies that operate them, give us the same level of respect.”
Reed said that even if driverless cars proved to be safer than vehicles driven by humans, there would still be mistakes and accidents, which could be very different from the types of incidents humans had on the roads.
He said AI models would be constantly updated and improved based on real-life incidents.
“The crash that happens in Manchester on Monday should help to prevent a crash in Telford on Tuesday,” Reed added.
Sign up to Herald Premium Editor’s Picks, delivered straight to your inbox every Friday. Editor-in-Chief Murray Kirkness picks the week’s best features, interviews and investigations. Sign up for Herald Premium here.