All products featured on WIRED are independently selected by our editors. However, we may receive compensation from retailers and/or from purchases of products through these links.
Self-driving cabs on London streets by 2021. It’s a bold claim, but not quite what the headlines suggest. The reality? Something much more mundane, and perhaps even more useful: autonomous buses.
The news that minicab firm Addison Lee was teaming up University of Oxford spinout Oxbotica to develop driverless cars immediately caught the imagination of the UK press. “Addison Lee plans self-driving taxis by 2021” wrote the BBC. Sky News struck a similar tone: “Self-driving taxis to launch in London by 2021, Addison Lee says.” This website, too, got in on the hype, with the Addison Lee story leading the daily WIRED news round-up.
But if you take the time to ask, neither Oxbotica nor Addison Lee are promising a fully automated car, with no human safety driver in the front seat, picking you up from home and taking you absolutely anywhere in the capital you'd like. Indeed, Addison Lee isn't planning to use the technology for minicabs, but for shuttle buses and ride sharing, notes a company spokeswoman.
It is looking for full automation, though. And three years is a short time to not only fine-tune a challenging technology, but to get regulatory approval, ensure security, figure out a business model, and get the public on your side. Can it be done in three years? The answer depends on who you talk to — the optimistic experts we spoke to said it was feasible given the pace of innovation, while the cynics just laughed — but also what you mean by driverless cars.
That full level of automation — level 5 on the driverless car scale — isn't likely in three years, says Graham Parkhurst, professor at the University of West England and the director of its Centre for Transport and Society. "Operating in completely automated mode at level 5 on the public road implies much better-than-human reliability in detection, identification, and responsive operation in the full range of road environments and conditions," says Parkhurst.
"That would include snow, fog, and heaving rain conditions on motorways, country lanes and busy, complex urban streets. It is not just about safe operation but also efficient operation — when it is safe to ignore movements that could theoretically be in conflict with the progress of the vehicle but in practice won’t be. Otherwise the roads, and junctions in particular, will grind to a halt."
And that's before adding in the complexities of London's roads — you could certainly pick an easier city to start with, notes Dr Shan Luo, an assistant professor in robotics at the University of Liverpool. "The behaviours of pedestrians on streets are hard to predict and bad human drivers can also cause trouble for a well-behaved AI driver," he says. "Narrow roads and complicated street layouts can be challenging for human drivers, will they be easier for AI drivers? Probably not."
On the other hand, Google's Waymo — which is much further ahead, having clocked up more than 10 million miles — is planning to launch driverless taxi services this year in Arizona. But here in the UK, projects are a bit further behind. So what is technically possible within three years?
Read more: WIRED's need-to-know guide to driverless car testing in the UK
The plan is to spend the next three years finding out, says Oxbotica CEO Graham Smith. On a technical front, Smith says that 12 to 15 Addison Lee vehicles will be loaded up with Oxbotica's sensors to map London's roads, capturing everything from roads to infrastructure to signs. "As these vehicles drive around London, just doing their daily business, they'll be recording data for us," Smith says, using cameras, LIDAR and radar.
The aim is to capture 250,000 miles of London, which should take less than a year, he says. That data will be used to power Oxbotica's navigational maps as well as for training autonomous systems. "Technically, we understand the challenges ahead of us," says Smith.
That may sound like a lot of miles to rack up, but it's a small slice of the tens of millions Waymo has slurped up. Smith thinks it's a sufficient starting point for a city like London, but stresses the cars would not be able to drive elsewhere in the UK. Indeed, don't expect the cars to even cover all of London by 2021. "We'll probably begin deployment in simpler areas that may be easier to control," Smith says. "As examples, things like campuses or airports are relatively simple compared to say Oxford Street or Regent Street in London. So we'll probably start to deploy in small pockets, and then look to join those together over many years."
Rather than central London, that means the less complicated suburbs could get such services first — but again, it's unclear what those services would be, and understanding that is part of the work of the next three years. Oxbotica has previously trialled slow driverless delivery vehicles to deliver groceries in Woolwich, but that estate was otherwise car-free. And separate projects in Milton Keynes and Coventry highlighted how each city has different challenges, that need to be addressed one by one.
And rather than minicabs, we're talking shuttle buses. Smith said Oxbotica is already in discussion with TfL to see where autonomous vehicles could help fill in missing gaps in public transport. "We're looking at specific and different types of use case to where autonomy could help improve local congestion and help address the needs of local people," says Smith. An Addison Lee spokeswoman noted: "The first stage – which has a 2021 target date – is likely to be in corporate shuttles, around airports or campuses to use to get to work, to study or get to the airport. For the foreseeable future, our premium services will continue to be driven by our 5000 drivers."
In other words, 2021 likely won't see on-demand driverless minicabs, but automated buses filling in transport gaps in certain areas or on specific routes. "I can see some implementations possible in some environments where we'll be able to take the safety driver out, assuming that the law has changed," Smith adds. It's a start, and could lead one day to full level-5 automation, but it's not what most of us imagine when we picture driverless cars.
Do the headlines generated by such vague project announcements risk disappointment? Perhaps. But they could be designed to soften legislators, suggests Christian Wolmar, transport expert and the author of Driverless Cars: On A Road to Nowhere. "It's so far-fetched that it seems to be only a PR thing," he says of the Oxbotica and Addison Lee plans. "I think it is part of the softening up process, basically saying that this technology is coming so you better start putting the fences up to stop people wandering into the street and stop cyclists from using certain corridors because we want those for driverless cars."
Wolmar points to an editorial written by Addison Lee's Catherine Hutt in the Evening Standard in August that called for London to adapt to driverless cars, rather than the other way around, saying "autonomous vehicles need a decluttered environment, with clear delineation between pedestrian and vehicle space". That said, an Addison Lee spokesperson said the Oxbotica sensors will be collecting "a lot of data that we use to train the software about London traffic, cyclists and pedestrians," so the company isn't expecting perfectly clear roadways.
Although creating software capable of navigating the hectic tumble of London’s streets is a challenge, it might be easier to build a fully automated car than declutter the city’s bike, taxi, and people filled roads. Siraj Shaikh, professor of systems security at the Institute for Future Transport and Cities at Coventry University, suggests that not-quite-full automation could be a useful tool for taxi drivers, perhaps with "privileged roads" linking up key transport hubs, such as Heathrow airport. Motorways already ban pedestrians and bicycles, after all, so dedicating one lane for only driverless cabs to use would let professional drivers hand over the wheel for part of the trip.
As serious as the technical concerns are, the legal, business and social challenges could prove larger hurdles. Even if a level-5 car exists, or we clear the streets of humans, regulators need to actually approve the technology, which is naturally likely once it's proven safe, but politicians are infamously behind the curve on innovation, and this is a major change. Next, a business model needs to be developed.
Oxbotica and Addison Lee are considering that as part of their work together, but regardless of whether this ends up being ride-sharing or more akin to existing cabs, Smith noted that the Velodyne sensors his company uses cost $100,000 (£77,0000 per car. "We need to see sensor prices come down substantially to make autonomy affordable," Smith says. And then there's passengers. Will they be happy to take a ride from a robot? Smith isn't sure public opinion will prove a challenge. "We've never met people that don't want a ride in the vehicle," he says. Of course, those that don't want a ride probably aren't hanging out around driverless cars companies.
With serious technical challenges perhaps matched by legal, business and social hurdles, cynicism is warranted around any self-driving car claims, but optimism and enthusiasm also aren't wholly misplaced. "The driverless vehicles are expected to reduce traffic and emissions by better planning the driving," notes Liverpool's Luo. "They will also increase the driving safety by preventing accidents caused by human fatigue driving or bad driving. Furthermore, the AI drivers will save more free time for us humans from driving." Such goals are worth working towards — even if they take longer than three years.
And, says Sally Epstein, machine learning expert at Cambridge Consultants, progress is happening quickly; it's not impossible that a real leap forward occurs in the next three years. "It's a bold, ambitious goal," she says. "And we wish them well. We're not saying that Addison Lee and Oxbotica will fail… we want it to succeed." Hopefully it does — even if success doesn't match our dreams of robot cars at our beck and call, but a smart system of semi-autonomous buses, instead.
This article was originally published by WIRED UK
