Dusseldorf The future of self-driving cars is not only emerging in Silicon Valley and Shanghai, but also in Israel. Not only the now world-famous Intel subsidiary Mobileye from Jerusalem is the world leader in corresponding sensors and self-learning systems. A new challenger is currently emerging in the port city of Tel Aviv, who wants to use superior technology to make autonomous driving better and cheaper.
Autobrains is the name of Igal Raichelgauz’s start-up. It uses self-learning artificial intelligence (AI) to improve autonomous systems. According to their own information, they need significantly less computing power and thus generate less heat – all decisive advantages. Because with less energy, electrically powered cars in particular can go much further.
In a new round of funding, the Israeli start-up has now raised $120 million from investors led by Temasek, Singapore’s sovereign wealth fund. They value Autobrains at an estimated $2.5 billion.
For the early investors, the entry has already paid off. In addition to the automotive supplier Continental, BMW iVentures, Knorr- Bremse and the Vietnamese electronics manufacturer Vinfast are also among the investors.
Top jobs of the day
Find the best jobs now and
be notified by email.
The Autobrains supervisory board is also prominent: the former Opel boss Karl-Thomas Neumann has invested a seven-figure sum in the start-up and heads the supervisory board. He is supposed to open the doors to the car industry for the start-up. Because despite well-known investors from the industry, no manufacturer has installed Autobrains systems to date.
Discussions and tests are currently underway with car manufacturers and suppliers. Autobrains is trying to do business in Germany. In addition to offices in Silicon Valley and Detroit, the Israeli company plans to open one in Berlin and Munich soon.
“Then the phone won’t be silent anymore”
For a successful market entry, the start-up from Israel has to prove that the AI can handle all traffic situations at least as well as systems from providers such as Mobileye, Renesas or Xilinx. “It’s not easy to get into a car,” says investor and board member Neumann. “If we can do it, the phone won’t be silent anymore, I’m absolutely convinced of that.”
Because Autobrains should make it much easier for an autonomous car to capture the environment. With previous methods such as deep learning, the AI is trained with photos in order to behave correctly in traffic. Objects such as trees, pedestrians or stop signs are shown to the neural network in as large a number as possible. The objects are equipped with “labels” so that they can be clearly identified – a complex job, mostly done by hand by people.
So far, this not only costs a lot of money, but also computing power. “The promise of fully autonomous driving seems five years away for good,” said the Autobrains boss a few months ago, “but not for Autobrains.”
The CEO of Autobrains has been researching AI for a long time and has registered around 50 patents in his name.
Raichelgauz himself is driving the development of his AI. Like many of his employees, he himself comes from the Israeli military and once served in the Israeli elite unit 8200, which listens to signals and cracks codes. Within a year, the number of employees has doubled to 120.
Together they are working on an AI that assigns so-called simple “signatures” to all objects: lines, colors or other structures in the image, without going into the last detail. “The AI does not need any learning,” says Neumann. “It recognizes a pedestrian by their signature, which is typical for pedestrians, but not by whether they are wearing a red or white jacket.”
Artificial intelligence should drive it like a human. He doesn’t concentrate on all possible scenarios all the time either. For example, if he drives on the freeway, roundabouts are “not an issue”. The Autobrains AI is supposed to emulate this with “agents”. A hierarchically structured neural network activates the required agents depending on the situation.
>> Read about this: These are the five hurdles on the way to autonomous driving
The effect: According to Autobrains, the AI only needs ten percent of the computing power compared to conventional neural networks. This is important because less heat is generated on the camera on the windshield or on the vehicle’s central computer – i.e. where the driving data is processed. This means that the systems need less cooling, which saves weight and increases the range, especially in electric cars.
According to Stefan Bratzel, Director of the Center of Automotive Management (CAM), unsupervised artificial intelligence, i.e. “unguided artificial intelligence (AI)”, is “promising”. The AI independently develops systems for object recognition – similar to human perception. Less training data is required for these algorithms and unknown objects are also recognized better, says Bratzel.
“The AI doesn’t need any learning – it recognizes pedestrians by their signature”
Kaspar Sage, Managing Partner of BMW iVentures, also considers the product to be “future-oriented” as a “scalable platform”. Because the algorithms require less processor power and can therefore also be used in inexpensive hardware, they are suitable for widespread use in the market. Sage tells the Handelsblatt that this is a competitive advantage over other providers who use training data.
The self-driving car business is expected to grow strongly: According to Roland Berger estimates, sales in the software market alone for cameras for autonomous driving up to level 2 will amount to 0.9 billion dollars and will grow to 3.4 billion dollars in five years Dollar.
Autobrains has set itself the goal of displacing the big competitor Mobileye in this lucrative market. According to competitor estimates, Mobileye currently controls up to 70 percent of the market for video recognition for autonomous driving. The Intel subsidiary will soon go public for $50 billion.
The former Opel boss discovered Autobrains while visiting various start-ups in Israel.
Investor Neumann is convinced that Autobrains can also challenge the market leader with its technology. The auto industry has a love-hate relationship with Mobileye. Their systems are reliable, but also more expensive because they “only work with Mobileye hardware”. A camera for self-driving functions therefore costs 100 to 120 euros. That’s cheaper.
Jan Mrosik, CEO of Knorr- Bremse, is also impressed by the technology. When the supplier joined Autobrains in November, he said in a press release: Autobrains’ self-learning AI technology has the potential to “decisively shape” the development of driver assistance systems (ADAS) and highly automated autonomous driving (HAD) in the commercial vehicle industry.
“Innovations in the automotive sector often come from start-ups that develop new, groundbreaking ideas,” agrees automotive expert Bratzel.
In the future, the technology will increasingly be taken up by important groups in the automotive and supplier industry. But the competition in the segment is fierce: Autobrains faces strong competition from established providers such as Mobileye or start-ups such as Helm.ai and operates in a field in which many companies “have been active and successful for years”.
More: The brain of Tesla – this is how the electric pioneer works on artificial intelligence