The moment I learnt to relax and trust the driverless car

Alex Kendall with a prototype driverless car at the Wayve headquarters in London

JOSHUA BRATT FOR THE SUNDAY TIMES

As the car ­— a Jaguar I-Pace — heads out of the Wayve headquarters near King’s Cross, a man called Darren is sitting in the driving seat. He will take control if something goes wrong but for now, we’re in the hands of an algorithm, six discreet cameras and a computer in the boot. (Wayve has special dispensation to test and develop its AI on public roads in the UK.)

When Kendall first took his AI vehicle for a spin around the suburbs of Cambridge in 2017, it was like being in the car with a nervous learner, he says. Today, it sets off with confidence.

Within a minute, something happens that I don’t immediately realise is exciting.

“Did you see that,” says Kendall, who has been described as a “once-in-a-generation brain” by his backers. “The car slowed a little as we went over a zebra crossing.”

The reason it slowed was because traffic on the other side of the road was bumper to bumper. It was possible that a pedestrian might nip between the bumpers and then step out into our lane. That anticipation of danger — a nuanced and complex human response — is what Kendall is excited about. “We didn’t program it to do that,” he says. “It learnt that from the data.”

Kendall’s decade of tinkering has been more complex than Winton’s (and not just because he doesn’t have six kids). At Auckland University, the New Zealander followed his father into engineering and then won a scholarship to Cambridge, where he spent his doctoral time pioneering deep-learning algorithms. The goal? To enable a robot to understand where it is, what’s happening around it and what will happen next via cameras alone. Since then, he has been applying his theories to the real world and an idea that’s simple, even if the science is mind-boggling.

If a robot can see and “understand” its situation, you can do away with the complex mapping and infrastructure used in most automated vehicles. In theory, it means Kendall’s robot can drive anywhere and cope with any situation. It learns from experience, just like we do.

We stop before a (poorly marked) no-stopping yellow box. Two vans are edging in ahead of us. We begin to move before they have cleared the box rather than waiting for the box to become completely clear (and risk other vehicles pushing in). “If you just wait and let objects move in front of you, you’re going to cause congestion,” says Kendall. “You’ve got to keep moving.”

Last week, the government passed the UK Automated Vehicles (AV) Act, which paves the way for autonomous vehicles on our roads by 2026. But what does this mean in practice? Are we actually going to see self-driving cars on the roads in two years?

In truth, automated “robotaxis” are already operating in California, within limited, mapped areas. If you’re a techno optimist like Edison, manual driving is doomed in the same way horses were a century ago.

Once Wayve has proved its deep-learning approach is safe, the company plans to move to fully hands-off, eyes-off autonomous driving. At that point, no one needs to sit in the driver’s seat. There doesn’t even need to be a driver’s seat. The robot can do the six-hour drive to Cornwall — you can sit in the back playing Yellow Car with the kids.

Wayve has special dispensation to test and develop its AI on public roads in the UK

TIM ANDREW

From there, the potential becomes quickly stunning. The average car in England is driven 4 per cent of the time and sits around clogging up the place for the remaining 96 per cent. If cars can drive themselves, car ownership could become a curiosity, the preserve of the die-hard petrolhead. The rest of us could summon cars only when we needed them. No more parking problems. We can all have front gardens again. Joy.

But will a true driverless car ever be able to cope with any situation anywhere? Can the current manifestation cope with the chaotic, scooter-ridden, bike-filled, pedestrian-spotted roads of inner London?

So far, so good. One pedestrian has even waved a thank you at Darren for letting her cross in front of us, even though Darren played no part in that decision. Our progress is serene. It’s also slow — in the 20mph limit, our top speed today is 18.387mph — and the software won’t allow us to speed. How would it cope in the back lanes of Shropshire or rush-hour Calcutta? How will it know that you can undertake in Maine but not on the M1? “The rules of physics are the same,” says Kendall. “Behaviours may be subtly different — the system will adapt.”

Kendall’s enthusiasm is infectious. He might be a nerd but he’s quite different from Mark Zuckerberg. His origin story involves surfing and mountain-biking in the wilds of New Zealand, not coding who’s hot and who’s not. He sounds comfortingly un-robotic when he talks.

Whether he can convince us to surrender our cars to an algorithm is a different matter. In a recent survey by the RAC, 70 per cent of respondents were concerned about the reliability of driverless cars’ software. The majority of us — ground down by a lifetime of computers saying no — are reluctant to put our lives in the hands of technology.

“The AV [act] is trying to get to the place where the technology is as good as if not better than a competent driver,” says Simon Williams, head of policy at the RAC. “If we get to that point, it would be welcome, but the most difficult thing will be the transition, where drivers are expected to take back control if something isn’t going according to plan.” The research, he says, is clear: it’s much harder to remain alert and focused — ready to jump in at a moment’s notice — if you’re not driving.

Until very recently, driverless technology just wasn’t good enough to withstand such criticism. With a lot of cameras, mapping and careful programming, you could just about get a car to drive itself around. But if anything unusual happened — a storm, a wonky road sign, a rogue e-scooter — the car would panic, stop and burst into digital tears.

California gave its first licence to deploy autonomous cars without safety drivers in 2020 — and it’s been a pretty chequered ride since. In San Francisco last spring, ten autonomous taxis came to a halt on one street because of “wireless bandwidth issues”. The resulting gridlock encompassed several blocks. Cruise, a subsidiary of General Motors, had its driverless permit suspended after several incidents including one in which one of its cars ran over and then stopped on top of a woman. She survived. A dog that ran out in front of a driverless taxi operated by the Google company Waymo did not. These and several other incidents have led to protests across the state.

As many as ten Cruise driverless cars stopped working in the busy North Beach neighbourhood of San Francisco last year

TERRY CHEA/AP

Think of these examples as AV 1.0 — cars with huge cameras and sensors operating with varying levels of success within a carefully mapped zone. Back in King’s Cross, however, I’m experiencing AV 2.0 — a more sophisticated and adaptable system.

We are approaching a bus as it disgorges its passengers. A lorry is coming the other way. If I were driving, I’d probably wait — an overtake is a bold move requiring skill to squeeze through the gap — but, surprisingly, the AI goes for it.

Then, just as we’re passing, the bus indicates to pull out. If our AI were a white van man, it would ignore the indicator and keep going. Instead, it backs off and tucks in behind the bus. Again, Kendall is excited. “You’ve got to keep moving but you’ve also got to be courteous,” he says.

Five years ago, a driving editor told me we’d never see fully automated driving because of the trolley problem. Imagine the scenario — a runaway trolley (or tram) is about to kill five people who are inexplicably standing on the tracks, but you can pull a lever and divert the tram to a track with only one person waiting to die. Do you pull the lever?

Your answer reveals what sort of a person you are — don’t tell me — but when you abrogate the responsibility for that decision to a robot, it becomes ethically tricky. Much like Wayve, Tesla is also now developing embodied AI in its self-driving cars. But if it emerged, for example, that Elon Musk had coded his software to prioritise the lives of customers over pedestrians, sales might go up — but so would the litigation.

In the UK, any incidents while the car is in self-drive mode will be the responsibility of the manufacturers. In the US, responsibility has become a matter for the courts.

Wayve claims that the technology is now becoming sophisticated enough to render the trolley problem academic: driverless cars would avoid putting themselves in a who-lives, who-dies situation in the first place. Kendall also points out that there are 1.2 million road deaths worldwide every year. Deaths and injuries on Britain’s roads cost the economy £8 billion a year. His goal is to reduce these devastating human and economic costs to near zero.

As we get back to the Wayve car park, I thank Darren for doing absolutely nothing and I decide I have no qualms at all about letting the robots drive. At every step of the car’s evolution — seatbelts, speed limits, drink-driving limits, all of it — people have protested too much. If and when one of Kendall’s cars runs over someone’s chihuahua, there will be disapproving headlines. But if my car can get up and take the kids to school, I will happily have a lie-in.