Mobility/News/

Wayve’s alternative autonomous driving approach gets $200m funding boost

UK startup has a different AI approach to self-driving cars, and aims to be first to deploy autonomous vehicles in 100 cities

By Maija Palmer

Alex Kendall, CEO of Wayve

Wayve, a London-based startup creating autonomous driving technology based on computer vision and machine learning, has raised a $200m Series B funding round to help get self-driving cars onto the road faster.

“This really signals the transition for us from being a very contrarian approach going against the rest of this multibillion-dollar industry to one that’s validated a lot of the core scientific claims and is now ready to commercialise at scale,” said Alex Kendall, cofounder and CEO.

Eclipse Ventures, a long-time supporter of Wayve, led the round, bringing in  D1 Capital Partners, Baillie Gifford, Moore Strategic Ventures, Linse Capital, Microsoft, Virgin, Compound and Balderton Capital. Ocado Group became a strategic investor in a previous round.

The funding round comes as many autonomous vehicle companies like Google’s sister company Waymo and General Motor’s Cruise have come under criticism for progressing slower than expected towards commercialising their technology. Autonomous vehicles have had huge amounts of money ploughed into them, with Waymo raising more than $5.5bn including a $2.5bn funding round last summer.

A number of early pioneers in self-driving cars, like Uber and Lyft, have sold their autonomous operations, worried that the technology is still a decade away from being ready.

A different approach

Wayve, which was founded in 2017 when New Zealand-born Kendall was still a research fellow at Cambridge University, has been building autonomous driving software from a different starting point to many competitors.

Companies like Waymo, Cruise and Aurora have relied heavily on technology like lidar and other sensors to survey the environment and then apply deep learning algorithms, drawing on huge datasets about driving scenarios and maps, to instruct the car how to move.

Wayve, on the other hand, has a relatively light, camera-based hardware element but relies more heavily on advances in machine learning. It uses a combination of advanced computer vision software — Kendall was part of groundbreaking research in this while at Cambridge — coupled with the kind of reinforcement learning that allowed computers to learn to play games like Go (if you want to understand the difference between deep learning and reinforcement learning, this is a good article to start with).

“We can enter new markets at a cost point that is far below the traditional stack and that’s why we’re going to be the first to be in 100 cities”

“The traditional autonomous driving stack has a very complex sensor setup with 30 cameras and other sensors. They use machine learning to understand the world but the way they behave and move is all hand-coded using rules and preset high-definition maps that tell the car where to drive,” Kendall tells Sifted.

Systems like these can get confused by situations they don’t recognise — which was the cause of the 2018 crash, in which a self-driving Uber killed a woman pushing a bicycle because it had mislabelled her as an unknown object, then a car and finally a bicycle.

Of course, it is entirely possible that a Wayve vehicle could react wrongly to a situation, too. But the system is better geared up for learning from unusual edge cases, as it doesn’t need huge datasets for training.

Faster and cheaper to deploy?

Wayve’s system first learns by imitating human drivers — it is currently installed in the fleets of delivery vans from DPD, Ocado and Asda, gathering information about driving. Wayve will then also give feedback and interventions to fine tune that behaviour.

Once the system has learnt to drive in one city, the autonomous vehicle can drive in any city with minimal new inputs, something that Wayve demonstrated last autumn, when its vehicles, trained in London, were able to drive in five other UK cities (Cambridge, Coventry, Manchester, Liverpool and Leeds) that they had never seen before.

“The car didn’t have a high-definition map and had never seen the cities before, but it drove through traffic light scenarios there because it learned that core behaviour,” says Kendall.

This approach, Wayve says, will make it cheaper and faster to get the technology into use in multiple different cities. Navigating somewhere like Paris, where cars drive on the right, would require some new training, but less and less is required each time it goes into a new geography.

“We can enter new markets at a cost point that is far below the traditional stack and that’s why we’re going to be the first to be in 100 cities,” Kendall tells Sifted.

Are we finally moving towards autonomous cars driving on our roads?

Wayve didn’t give a date for when it was likely to be rolling out to those 100 cities. This year, however, it will be running autonomous delivery vehicles — with a human safety driver still on board — in joint pilot projects with supermarket groups Asda and Ocado.

Meanwhile, the UK is taking baby steps towards allowing autonomous driving onto the roads. Last year is allowed hands-free driving using automated lane-keeping systems in some circumstances, and there is an ongoing consultation on allowing self-driving cars more broadly.

Maija Palmer is Sifted’s innovation editor. She covers deeptech and corporate innovation, and tweets from @maijapalmer

2
Join the conversation

avatar
  Subscribe  
newest oldest
Notify of
Commenter
Commenter

How can you write an article about autonomous driving with no mention of Tesla??

Commenter2
Commenter2

In particular because Tesla also doesn’t use Lidar and HD maps, relying on computer vision instead – just like Wayve. It’s just a different architecture and set of algorithms.