The British self-driving car company Wayve is claiming a big breakthrough in tackling one of the industry’s toughest challenges — how to “generalise” robotics software so it can drive any vehicle anywhere.
The London-based startup says it's successfully used software trained on its autonomous Jaguar I-PACE passenger car fleet to drive light commercial Maxus e9 vans, which have different dimensions, wheelbase and driving dynamics.
“It’s a complete paradigm shift for autonomy. We’ve got to have intelligent machines that can adapt to the really dynamic and complex worlds that we live in,” Alex Kendall, Wayve’s chief executive, tells Sifted. “This is what lets us drive everywhere.”
But the big hurdle now will be rolling out the tech.
Kendall is concerned that rumoured delays to the British government’s Future of Transport bill may frustrate efforts to create an internationally recognised safety regime to verify and permit the speedy deployment of new breakthroughs in tech.
“It’s incredibly important the UK does bring legislation forward for autonomous driving,” says Kendall. “We remain confident that the UK government sees this as a priority, sees this as an opportunity to have an independent global company come out of the UK and really define what is going to be a trillion-dollar market.”
Developing vehicle-agnostic technology
When Kendall founded Wayve in 2017, he says that rival autonomous car companies laughed at his approach, which had been discarded by most of the industry.
Wayve’s software relies on vision technology underpinned by sophisticated reinforcement learning. But other big autonomous car companies, such as Waymo and Cruise, include lidar and radar systems as well. In theory, that can increase their reliability. But including as many as 30 different sensors also adds cost and complexity.
I was blown away in the backseat. I was really impressed by the intelligence of the system
By contrast, the New Zealand-born computer vision expert and former research fellow at Cambridge was convinced that self-driving systems would have to learn to adapt to different situations — just as humans do — if they were to operate everywhere.
Kendall says that every journey a vehicle makes is different depending on weather, lighting, driving conditions, roadworks and interactions with other users, making it difficult to rely on pre-programmed rules. “You’re never going to see the same thing twice. This is the main argument why generalisation is important,” he says.
One incident, which highlights the software’s flexibility, struck him a few weeks ago when one of Wayve’s vehicles stopped at a temporary traffic light surrounded by red cones, forcing cars to drive on the wrong side of the road. Even though the traffic lights stayed red, the vehicle responded to a road worker waving through the traffic. “I was blown away in the backseat. I was really impressed by the intelligence of the system.”
Wayve is now pioneering this vehicle-agnostic AV2.0 technology with the aim of becoming the first company to launch self-driving cars in 100 cities.
After trialling its autonomous cars in central London, Wayve has extended its testing to Liverpool, Cambridge, Manchester, Coventry and Leeds. It has also trained its software on multiple simulations modelling edge cases. With only 80 hours of specific testing on vans, Wayve’s software was able to drive these vehicles safely. They will soon be tested by Wayve’s partners Ocado and Asda to help solve the “last mile” delivery problem.
Ocado plans tests of Wayve’s autonomous vans
The grocery delivery company Ocado, which has invested £10m in Wayve, is now planning to conduct small-scale tests on autonomous vans over the next 12 to 18 months near its head office in Hertfordshire. If successful, the company believes it can build an economic case for their commercial use within five to six years, assuming that enabling legislation is in place and that consumers are prepared to collect their groceries from the vans.
There is a sizeable, if not existential, threat [to a company] if the technology is not deployed in ways that do not have the confidence of the public
Alex Harvey, chief of advanced technology at Ocado, says that Wayve’s approach is “very exciting”. But he stresses that Ocado will be extremely cautious in adopting autonomous technology for regular deliveries. “There is a sizeable, if not existential, threat [to a company] if the technology is not deployed in ways that do not have the confidence of the public,” he says.
Tom Bruls, CTO of computer vision startup Synativ, says that integrated “behavioural cloning models” — such as Wayve’s — can be a lot more efficient and cost-effective than other autonomy stacks because “you do not have to hand-engineer every system for every situation”. “What they are doing is very cool and they are solving a big problem in autonomy,” he says.
But Bruls argues that the downside of such models is that it might be more difficult to identify faults when things go wrong. “The more you move towards a one-model-does-everything approach it is not so explainable,” he says. “That may make it harder for the regulators to accept.”
To date, Wayve has raised $258m of financing. Its backers include Firstminute Capital, Balderton and Baillie Gifford. It has also raised funds from Eclipse Ventures, D1 Capital Partners, Moore Strategic Ventures, Linse Capital, Microsoft, Virgin, Compound and Ocado.
For the moment, Kendall is determined that Wayve should be the first autonomous vehicle company to operate in 100 cities. But longer term, he is focused on the many other future uses of embodied robotic intelligence. “For me, the real value of AI comes when we embody it in the real world. And at Wayve, we’re really building a giant data corpus and the learning system to build foundation models for embodied intelligence,” he says.
John Thornhill is Sifted’s editorial director and cofounder. He is also innovation editor of the Financial Times, and tweets from @johnthornhillft.
Looking for digestible insights into the autonomous vehicles sector? Sifted’s Pro Briefing on the industry will get you up to speed fast on what you need to know. Check out what Pro membership can offer you here.