Projects like Google’s Waymo, Uber, Cruise and Aurora are developing autonomous vehicles by throwing engineers at the problem, basing most of their platforms on rule-based systems that try to pre-empt and deal with every edge case, whilst also peppering the cars with more sensors to capture more data. This can work in relatively controlled environments but has the drawback of not being able to flexibly adapt in real-time to fast-changing situations.
Despite all this investment and many years of development, no one has yet been able to launch a commercial autonomous car service. It’s just very hard to hand-engineer. What’s required is not more eyes but better coordination. The simple answer — as it is to almost everything these days — would be to throw AI at the problem, and that’s what many startups, which lack the engineering and hardware muscle of the big players, are trying to do.
Now they say they do, and the results are not only fascinating but might also be genuinely innovative.
In fact, they are claiming a “world first” in demonstrating that a car working on their machine-learning platform can drive on roads it’s never seen before during training, and without an HD map of its environment. Other systems, like Waymo’s, rely on maps and rules to drive. Theirs, says Wayve, does not.
Additionally, it’s also revealed that it’s been testing its platform on the Jaguar I-PACE SUV, which won the 2019 European Car of the Year. Interestingly, this is also a car which has been used by Waymo in some tests.
Alex Kendall, Co-Founder & CTO tells me: “Our cars learn to drive from data with machine learning. Every time a safety driver intervenes and takes over, the car learns to drive better. We don’t tell the car how to drive, rather it learns to drive from experience, example and feedback, just like a human. This is more safe and scalable than any other approach today.”
Emerging out of research from the University of Cambridge, Wayve has already undertaken extensive testing on public UK roads.
Kendall says: “The traditional approach used by all our competitors relies on HD-maps, expensive sensor suites and hand-coded rules that tell the car how to drive. We have built a system that learns end-to-end with machine learning. It is the first in the world to drive on urban roads it has never been on before. It uses compute/sensors which cost less than 10% of competitors.”
Bold claims. But a video today revealed on their site shows their system driving on public roads in Cambridge, UK, driving on roads it has never been on before using a sat-nav route map and basic cameras.
“We don’t tell the car how to drive with hand-coded rules: everything is learned from data. This allows us to navigate complex, narrow urban European streets for the first time. End-to-end deep learning,” says Kendall.
“Our model learns both lateral and longitudinal control (steering and acceleration) of the vehicle with end-to-end deep learning. We propagate uncertainty throughout the model. This allows us to learn features from the input data which are most relevant for control, making computation very efficient. In fact, everything operates on the equivalent of a modern laptop computer. This massively reduces our sensor & compute cost (and power requirements) to less than 10% of traditional approaches,” he says.
Assuming other independent observers can confirm these claims, it looks like a UK startup just leap-frogged the entire autonomous car space.
For now Wayve is being coy about it’s investors, saying only that Professor Zoubin Ghahramani, Chief Scientist of Uber, is an investor.