This Startup Is Betting on AI for Autonomous Vehicles

• UK-based autonomous developer Wayve has opened an office in the US and plans to launch testing in the Bay Area, having secured $1.05 billion in funding earlier this year.

• The company takes an AI-centric approach to autonomous driving, relying on a system that does not require detailed 3D mapping of every environment and can work with different types of sensors.

• A number of autonomous tech developers, including Tesla, now back a vision-only approach to sensor inputs while relying on neural networks, as opposed to code, for decision making.

A decade ago higher levels of autonomous driving tech were largely based on concrete rules, with systems following programming code in interpreting input from various sensors on vehicles.

But as developers began to make gains in artificial intelligence over the past few years, it has become possible to make autonomous driving software rely on AI decision-making, trained by years worth of video data.

Tesla has recently embraced this approach as well in its ever-evolving FSD Beta, betting on AI on its path to more capable autonomous vehicles.

And an AI-based approach is what UK-based startup Wayve has in mind as well, as it launches testing in the Bay Area after opening its first site in the US.

Why look to AI, after years of industry research into autonomous vehicles that rely on code?

"As industry hype begins to dissipate, the challenge is becoming even clearer: The classic robotics approach—known as AV1.0, which relies on complex sensors, labor-intensive HD mapping, and hand-coded rules—is proving to be increasingly cumbersome and cost-prohibitive to build," said Alex Kendall, Wayve co-founder and CEO.

The company has been working on what it calls AV2.0, which relies on self-learning in the process of adapting to new driving environments in any country in the world.

It's an approach that bypasses the arduous process of 3D mapping every street down to a fraction of an inch—until recently seen as a necessary part of geofenced SAE Level 4 operations. Wayve instead relies on a camera-only approach to navigating streets, but also open to other sensor inputs including lidar and radar.

Wayve has been a believer of this end-to-end AI approach since launching in 2017.

The advantages, according to the company, also include the ability to operate as part of a variety of vehicles with different sensors and hardware, in addition to being suitable for use in different countries without relying on detailed maps of the road infrastructure, or what the company calls generalization.

"We are already training on diverse data from over 15 countries and are developing an AI Driver product that learns from diverse driving cultures, enhancing its ability to operate globally," Kendall added.

Wayve now plans to begin testing its vehicles in the San Francisco Bay Area, after opening an office in Sunnyvale.

But at first, Wayve plans to test its more basic Advanced Driver Assistance Systems (ADAS) systems on US roads before moving on to higher levels of autonomous tech.

The company isn't a robotaxi fleet operator per se, in the same way that Waymo is structured. So we won't see Wayve-branded robotaxis on US roads that can be hailed via an app. Rather, the startup is aimed to offering its software directly to automakers, starting with SAE Level 2 and higher.

Even though machine learning has been a part of autonomous research for quite some time, it is becoming increasingly clear that the path to ubiquitous Level 4 driving will include some form of AI, even as developers might take different approaches to achieving reliable Level 4 operations.

At the moment Wayve has gathered quite a few believers, including Microsoft and Nvidia, landing $1.05 billion in the latest investment round earlier this year. But amid the expansion of robotaxi tech by rivals, at some point it will also need to find major buyers for its software.

One of the major hurdles for autonomous tech in the next few years, whether it's AI-driven or not, may well prove to be regulatory rather than technical in nature.

Will Level 4 robotaxis achieve significant market share by 2030, or will this process take longer? Let us know what you think in the comments below.