Autobrains CEO Igal Raichelgauz has so far secured 140 million dollars in investments.Autobrains
When thinking of autonomous driving and Israeli companies, most people probably think of Mobileye first. In addition, there is a second company, Autobrains, whose CEO reveals in an interview how his technology stands out from the competition.
Mr. Raichelgauz, at IZB 2024 you presented Autobrains' new product line for assisted and autonomous driving called Skills. The technology consists of a multitude of neural networks and is based on the company's proprietary Liquid AI technology. How does this technology address the challenges that existing systems have with regard to edge cases?
One of the main problems with today's systems is that they are optimized for average scenarios. Typically, a large neural network is used, which is trained with a huge amount of data to optimize the average across the entire training set. That's the problem. Liquid AI, on the other hand, is much more adaptable. Instead of optimizing a complete neural network for an average, it breaks down the problem into smaller parts and optimizes them separately. This reduces complexity. The reason we call it "liquid" is that information flows through the system like a stream and is directed to different areas. These areas work in isolation and independently of each other. By breaking down into smaller problems, we can address the "long tail" of edge cases without getting stuck in average cases.
When did you start developing this technology?
The technology is already relatively mature. We started the company about five years ago, and the technology was mature at that time, but not yet for the automotive sector. The company was founded to apply proven technologies from other fields such as medical imaging or security screening to the automotive sector. Our first project was the development of an ADAS system, and now we are scaling towards full autonomy. We have been working on it for five years.
Advertisement
How do the Skills products differ, for example, from the Mobileye approach?
Mobileye's approach is based on supervised learning methods. They use data for various scenarios, such as pedestrians and vehicles, which have been manually labeled and are permanently retrieved. This makes the system relatively expensive. Mobileye uses different modules, such as the perception module for perception, the planning module, and the control module. Liquid AI, on the other hand, is much more adaptive. It does not constantly use all resources for every scenario but activates the appropriate resources, making it much less computationally intensive. This significantly reduces costs. We also do not need a large number of people to manually label data. Our technology is capable of learning independently what pedestrians, cars, or traffic lights are in the images. This allows us to make labeling much more efficient, significantly reducing the overall costs for computing power as well as for training and validation.
You also said that your technology is better than Tesla's. Can you explain your confidence?
Tesla follows an approach based on a vision-only solution to develop an autonomous vehicle that can drive anywhere. Tesla is a leader in vision technology, and they have a clear advantage through their fleet and infrastructure. But Tesla still uses a single huge neural network, and to cover edge cases, they have to invest enormous resources. Our technology, on the other hand, does not require exponential resources to achieve the same results. We come with a more elegant solution that is more efficient.
How far along are you in terms of collaboration with OEMs? Are OEMs already using your technology?
Our product will hit the road in China by the end of this year and in Europe in the first half of next year.
Can you reveal which OEMs are involved?
No, not yet. We will announce the information once it is released.
Is it easier to integrate your technology into existing systems?
About the person:
Igal Raichelgauz, founder and CEO of Autobrains, started his career at the military elite intelligence service before co-founding Figment, a messaging platform. Later, he became the CTO of LCB, a speech recognition company. In 2007, Igal co-founded Cortica, an AI company known for its self-learning AI technology for visual perception. At Cortica, he was instrumental in founding seven AI companies in various fields, including Autobrains in the field of autonomous driving.
Yes, our approach is software-based and hardware-agnostic. This means that we do not need our own chips or special hardware. We can rely on existing chips and hardware, which makes integration much easier. We are also sensor-agnostic, meaning we can seamlessly integrate into existing systems. Our solution fits very well with software-defined vehicles, where software updates can be carried out over the air interface (Over-the-Air) without requiring hardware changes.
What SAE level does your technology currently master?
We start at Level 2+ and make it extremely cost-efficient. We then scale up to Level 4.
Is the technology also applicable for robotaxis?
No, we do not focus on robotaxis, as they operate in a geofencing-based environment, which is less challenging. We focus more on vehicles that can drive anywhere without supervision, as Tesla aims to do.
In Germany, there is a strong emphasis on safety and intensive testing procedures before vehicles are approved. Do you think it is more difficult to implement such technologies in Germany?
Yes, in Germany the bar is set very high in terms of quality. In China, the focus is more on speed, while in Germany quality is the priority. Our goal is to meet both requirements.
