Wayve CEO shares his key ingredients for scaling autonomous driving tech 

Co-founder and CEO of Wayve Alex Kendall sees promise in bringing the technology of his autonomous vehicle startup on the market. That is, if Wayve sticks to its strategy to ensure that the automated driving software is cheap to run, hardware-agent and can be applied to advanced driver systems, robotaxis and even robotics.

The strategy, which Kendall was recorded during the GTC conference of Nvidia, starts with an end-to-end data-driven learning approach. This means that what the system “sees” through different sensors (such as cameras) translates directly into how it drives (such as deciding to brake or turn left). Moreover, this means that the system does not have to rely on HD cards or on rules-based software, as previous versions of AV Tech have.

The approach has attracted investors. Wayve, which was launched in 2017 and has collected more than $ 1.3 billion in the last two years, is planning to licensed his self-driving software on automotive and fleet partners, such as Uber.

The company has not yet announced automotive partnerships, but a spokesperson told WAN that Wayve is in “strong discussions” with several OEMs to integrate its software into a series of different vehicle types.

The cheap-on-run software epitch is crucial to conquer those deals.

Kendall said that OEMs place the advanced driver system (ADAS) of Wayve in new production vehicles, do not have to invest in extra hardware because the technology can work with existing sensors, which usually consist of surround cameras and some radar.

Wayve is also ‘silicaagnostic’, which means that it can perform its software on every GPU that already have its OEM partners in their vehicles, according to Kendall. However, the current development fleet of the startup uses the Orin-System-A-A-Chip of Nvidia.

“Coming in Adas is really crucial [Level] 4, “Kendall said on stage on Wednesday.

(A running system of level 4 means that it can navigate itself in an environment – under certain circumstances – without a human being.)

Wayve is planning to commercialize his system at an ADAS level first. So, the startup has designed the AI ​​driver to work without Lidar -the light detection and varying radar that measures that distance using laser light to generate a very accurate 3D card of the world, which most companies develop that level 4 technology as an essential sensor.

Wayve’s approach to autonomy is comparable to that of Tesla, that is Also work on an end-to-end deeper model to provide its system with power and to continuously improve its self-driving software. As Tesla tries to do, Wayve hopes to use a widespread rollout of Ada’s to collect data that helps his system to achieve complete autonomy. (Tesla’s “full self-driving” software can perform some automated driving tasks, but is not completely autonomous. Although the company wants to launch a robotaxi service this summer.)

One of the most important differences between the approaches of Wayve and Tesla from a technical point of view is that Tesla only trusts on cameras, while Wayve likes to be Lidar of the full autonomy in the short term.

“Longer term there is certainly an opportunity if you build up reliability and the possibility to validate a scale level to reduce that [sensor suite] Furthermore, “said Kendall.” It depends on the producer experience you want. Do you want the car to drive faster faster? Then you might want other sensors [like lidar]. But if you are willing that the AI ​​understands the limitations of cameras and is therefore defensive and conservative? Our AI can learn that. ‘

Kendall also teased Gaia-2, the newest generative world model in Wayve, tailored to autonomous driving that trains its driver on enormous quantities of both Real-World and synthetic data over a wide range of tasks. The model processes video, text and other promotions, which Kendall says it enables the AI ​​driver of Wayve to be more adaptive and human in driving behavior.

“What is really exciting for me is the human -like driving behavior that you see arise,” said Kendall. “Of course there is no hand -coded behavior. We don’t tell the car how to behave.

Wayve shares a similar philosophy such as autonomous truck startup Waabi, which also strives for an end-to-end learning system. Both companies have emphasized the scaling of data-driven AI models that can generalize in different driving environments, and both rely on generative AI simulators to test and train their technology.