Wayve Alex Kendall’s co -founder and chief executive sees the promise of bringing the technology of the autonomous vehicle to the market. That is, if Wayve sticks to her strategy to ensure that automated driving software is cheap to run, the material agnostic and can be applied to advanced driver assist systems, robotia and even robotics.
The strategy, which Kendall wrote during NVIDIA’s GTC conference, begins with a data -based learning approach. This means that what the system “sees” through a variety of sensors (such as cameras) translates directly to how it drives (as it decides to brake or turn left). In addition, this means that the system does not need to be based on HD maps or rules -based software, such as previous versions of AV Tech.
The approach has attracted investors. Wayve, which began in 2017 and increased more than $ 1.3 billion over the last two years, plans to grant license to use self-guidance software to car and fleet partners, such as Uber.
The company has not yet announced any car partnership, but one spokesman told TechCrunch that Wayve is in “strong discussions” with many OEMs to integrate its software into a series of different types of vehicles.
Its cheap software step is vital to completing these agreements.
Kendall said that, by placing the advanced Wayve driver assist system in new production vehicles, it does not need to invest anything in additional material, because technology can operate with existing sensors, which usually consist of surround cameras and a radar.
Wayve is also “Pyritina-Identification”, which means he can execute his software in whatever GPU OEM partners already have their vehicles, according to Kendall. However, the current starting fleet of the start uses the Orin System-on-a-Chip of Nvidia.
“Entrance to Adas is really critical because it allows you to create a viable business, create a scale distribution, and obtain exposure to data so that you can train the system until [Level] 4, “Kendall said on Wednesday.
(A level 4 driving system means that it can navigate in an environment in itself – under certain circumstances – without the need for a person to intervene.)
Wayve plans to first commercialize his system at Adas level. Thus, the start designed the AI driver to operate without Lidar – the light and ripple detection radar that measures the distance using laser light to create an extremely accurate three -dimensional map of the world, which most companies developing technology 4 consider to be a basic sensor.
Wayve’s approach to autonomy is similar to Tesla’s, which is He also works on a deep-to-end-to-end learning model to feed his system and constantly improve self-guidance software. As Tesla tries to do, Wayve hopes to take advantage of a widespread Adas development to collect data that will help her system reach complete autonomy. (Tesla’s full self-guiding software can perform some automated driving tasks, but not fully autonomous.
One of the main differences between Wayve and Tesla’s approaches is that Tesla is based only on cameras, while Wayve is pleased to incorporate Lidar to reach short -term complete autonomy.
“Long term. There is definitely an opportunity when you create reliability and ability to validate a level of scale to shrink this [sensor suite] Further, “Kendall said.” It depends on the experience of the product you want. Do you want the car to drive faster through fog? Then you might want other sensors [like lidar]. But if you are willing for AI to understand the limitations of the cameras and be defensive and conservative as a result? Our AI can learn this. ”
Kendall also drives GAIA-2, Wayve’s latest genetic model, adapted to autonomous driving that trains its driver in huge quantities of both real and synthetic data in a wide range of work. The model edits the video, the text and other actions together, which Kendall says it allows Wayve’s AI driver to be more adaptive and human in driving.
“What is really exciting for me is the human driving behavior you see,” Kendall said. “Of course, there is no handmade behavior, we do not say in the car how to behave, there is no HD infrastructure or maps, but instead, emerging behavior is guided and allows for driving behavior deals with very complex and different scenarios, including scenarios that have never seen during training.”
Wayve shares a similar philosophy to the autonomous start of Waabi trucks, which also seeks an end -to -end learning system. Both companies have emphasized the escalation of AI models based on data that can generalize in different driving environments and both rely on AI genetic simulators to test and train their technology.
