Can AI help make better drivers out of us all?
▶ With human error playing a huge role in deaths resulting from road accidents, manufacturers are looking to self-driving cars. But, Rhodri Marsden asks, can their automated systems make up for a driver’s poor judgment?
Ever since the very first fatality was recorded in 1869, cars have posed a mortal threat to pedestrians. In 2013 the World Health Organisation estimated the annual number of pedestrian deaths on the roads to be more than a quarter of a million. In some countries, the number of fatalities is rising. Experts aren’t certain of the reasons; it could be down to the increasing popularity of powerful SUVs, or the distracting nature of in-car entertainment systems.
But with 94 per cent of the traffic fatalities in the US in 2016 attributable to human error, according to the National Highway Traffic Safety Admistration, automated systems are now attempting to make up for our poor judgment. Pedestrian detection and automated braking systems are combining to stop cars in an emergency, but their efficacy is proving hard to measure. Some tests describe them as “ineffective”. Others praise their “superior” performance. So how good are they, exactly? Are they making pedestrians safer, or are they making drivers blasé about danger?
Underlying all of these issues is the manufacturer’s profit. Equipping cars with the best PCAM equipment is expensive
Our perception of “intelligent cars” is conflicted. On one hand, we’re aware that self-driving vehicles are being tested around the world, and their very existence seems to represent some kind of modern miracle. On the other, any errors they make are pounced upon as evidence of their poor judgment. They’re rightly held up to a much higher standard than human beings, but they still haven’t reached the necessary level they need to be at, says Michael Clamann, an engineer at the Highway Safety Research Centre, based at the University of North Carolina in the US. “Technology for self-driving cars is very immature, and we have a long way to go before that technology is ready.”
The death of Elaine Herzberg provides distressing proof. In 2018, she became the first person to be killed by an automated car when she was hit by a Uber self-driving test vehicle. Last week, the US National Transportation Safety Board found that the car’s pedestrian detection system had failed to identify her as such. But while the Uber car relied on a human driver to take control if the automated system failed, standard cars equipped with such systems work the other way around. Drivers are held responsible; computers merely offer backup. And as a backup system, rather than a controlling system, the evidence is clear; they reduce collisions and save lives.
The systems have two components. There’s advanced emergency braking (AEB), which detects any kind of frontal collision – including with other vehicles – and brings the car to a halt. It works well at lower speeds and it’s set to become standard – 20 manufacturers have promised to incorporate AEB into all new cars by 2022, and in the US, it will be a legal requirement from that date. Then there’s pedestrian crash avoidance mitigation, or PCAM, which uses various sensors to recognise people in the road and trigger the AEB accordingly. Cameras, light detection and radar are deployed in various combinations depending on the model of the car. Their task: to capture data and compare it with millions of images to determine the level of danger to both car and pedestrian.
PCAM systems have a fearsomely complex task, given the limitless number of elements that factor into an accident, including the weather, the pedestrian’s clothing, the time of day and much else besides. In recent weeks, two bodies, the American Automobile Association (AAA) and the Insurance Institute for Highway Safety (IIHS), tested a number of PCAM systems in cars manufactured by Honda, Toyota, Tesla, Audi, Nissan, Subaru and many others. Their conclusions make for confusing reading. In one AAA test, where they simulated a child running out from behind parked cars, 89 per cent of the systems failed at 32 kilometres per hour, and all of them failed at 48kmh. Meanwhile, six of the cars in the IIHS study received a “superior” rating – including the Volvo S60 and the Mercedes-Benz C-Class – but some still hit a pedestrian dummy. In addition, the best performing cars weren’t always the luxury models. There seemed to be only one clear takeaway from both studies: these systems are far from failsafe.
“You can programme automatic braking to work under certain conditions, but there’s always going to be exceptions to the rule,” says Clamann. “No crash avoidance technology is designed to address every possible crash scenario – for example if a pedestrian is crossing from between a bunch of bushes, or if there’s a lot of visual noise on the side of the road that makes them harder to see. It’s really important to recognise that automatic emergency braking is not a replacement for human attention. If people became dependent on these systems instead of trying to watch for pedestrians, that would be really bad.”
PCAM technology will continue to improve, as will safety features outside the car – improved lighting, smart infrastructure, road design and speed limits. But one of the biggest challenges, says Clamann, is how to communicate the limitations and functions of these computerised systems to drivers in a way that inspires trust, but not too much trust. “You can’t say they work perfectly all the time, but you also need to be careful about saying that they never work,” says Clamann. “If drivers over-trust the system, they could end up crashing in the event of it failing. If they under-trust it, they might just turn it off, at which point they’ve lost all of the benefits because they don’t even have it working in the background. Manufacturers need to get the consumer into that middle ground, where they understand what the limitations are, while also being able to properly interpret alarms and get the most out of it. Unfortunately we’re not quite there yet.”
Underlying all of these issues is the manufacturer’s profit. Equipping cars with the best PCAM equipment is expensive. And overstating the capabilities of safety systems could be counterproductive; for customers to understand them is hardly the best advertisement for a car. The safety of modern cars, it seems, will continue to be inextricably tied to the psychology of the modern driver, but the message to that driver is clear: concentrate on the road. In the meantime, technology will do its best to help you, but don’t expect it to replace you. People’s lives are at stake.
Above, the Volvo S60 scored a superior rating for its pedestrian crash avoidance mitigation systems; left, Arnold Schwarzenegger takes a driverless taxi in the 1990 film ‘Total Recall’