Wayve co-founder and CEO Alex Kendall sees his commitment to bringing the technology of self-driving car startups to the market. This means that if Wayve is stuck on a strategy to ensure automated driving software, it can be agnostic to hardware and applied to advanced driver assistance systems, Robotaxis, and even robotics.
The strategies Kendall laid out during NVIDIA's GTC meetings begin with an end-to-end, data-driven learning approach. This means that what the system “sees” through various sensors (such as cameras) translates directly into how it drives (such as braking or turning left). Furthermore, it means that the system does not need to rely on HD maps or rule-based software, like in previous versions of AV Tech.
This approach attracted investors. Released in 2017 and raised more than $1.3 billion over the past two years, Wayve plans to license self-driving software such as Uber to its autonomous driving partners.
The company has not yet announced its automotive partnership, but a spokesman told TechCrunch that Wayve is in “strong debate” with multiple OEMs and that it has integrated the software into various vehicle types.
Cheap run software pitches are important for clinching these transactions.
Kendall doesn't need to invest anything in additional hardware as OEMS put Wayve's advanced driver assistance system (ADA) into new production vehicles, as the technology can work with existing sensors, usually consisting of surround cameras and radar.
According to Kendall, Wayve can also be “silicon dependent.” However, the startup's current development fleet uses Nvidia's Orin System-on-chip.
“Admission to ADAS is really important because it allows you to build a sustainable business, build distributions at scale, gain data exposure and train your system. [Level] 4, Kendall said on stage Wednesday.
(Level 4 driving systems mean that they can navigate the environment alone under certain conditions without the need for human intervention.)
Wayve will first commercialize the system at the ADAS level. Therefore, the startup designed the AI driver to run without LIDAR. Measure distance using laser light to measure distance to produce a highly accurate 3D map of the world.
Wayve's approach to autonomy is similar to Tesla, and Tesla is also working on an end-to-end deep learning model, enhancing its systems and continuously improving its autonomous driving software. As Tesla is about to do it, Wayve wants to leverage the widespread deployment of ADA to gather data that will help the system reach full autonomy. (Tesla's “fully autonomous driving” software can perform some automated driving tasks, but it is not completely autonomous. However, the company aims to launch its Robotaxi service this summer.)
One of the main differences between Wayve and Tesla's approaches is one of the approaches from a technology perspective. Tesla relies solely on camera, while Wayve is to incorporate Lidar and reach full short-term autonomy.
“In the long run, there certainly is an opportunity when building the ability to examine the level of reliability and scale and reduce it. [sensor suite] Additionally, Kendall said. “It depends on the product experience you want. Want to drive your car faster in the fog? Then you need other sensors [like lidar]. But what if AI understands the limitations of cameras and, as a result, wants to be defensive and conservative? Our AI can learn it. ”
Kendall also teased Gaia-2, Wayve's latest generation world model for autonomous driving that trains drivers in both real and synthetic data across a wide range of tasks. This model handles video, text, and other actions together. This allows Wayve's AI drivers to become more adaptive and human-like in their driving behavior, according to Kendall.
“What's really exciting for me is the human driving behavior you see,” Kendall said. “Of course, there are no hand-coded behaviors. They don't communicate behavior to the car. They don't have infrastructure or HD maps, but the emergency behavior is data-driven, allowing driving behaviors that handle very complex and diverse scenarios, including scenarios you've never seen before during training.”
Wayve shares a similar philosophy with autonomous trucking startup Waabi, which also pursues end-to-end learning systems. Both companies emphasize scaling data-driven AI models that can be generalized in a variety of driving environments, and both rely on test and training generation AI simulators.