May 28, 2022

AI has been the basis of many technological breakthroughs over the years, but one problem that remains to be solved is the possibility of self-driving: no matter how hard they try, the engineers have not yet created a practical platform. for a vehicle as good or better than a person and convinced regulators and the general public of its reliability. However, we are still seeing significant growth, and today Autobrains, one of the contenders for this niche, believes that with the help of a “machine learning” approach, it was possible to solve the 1% error problem typical for self-driving cars. . hardware agnostic (more on that below) announces even more funding to further develop its platform.

An Israeli startup raised $19 million, completing a $120 million Series C. The first tranche of this investment was made public in November 2021, and the general list of investors includes: Temasek, former strategic donors Continental and BMW i Ventures, as well as new donors Knorr-Brems AG and Winfast. As before, the company does not disclose its estimate, but in some context it is busy providing comparable numbers.

Israel-based Mobileye, which Autobrains CEO and founder Yigal Rachelgauz described as his company’s biggest competitor, filed confidentially for an IPO earlier this month (the owner of Intel would retain a stake in the floundering company if that happens). It is reported that Mobileye could be valued at around $50 billion if listed. Another Israeli self-driving car startup, Wave, raised $200 million in January, valued at about $1 billion.

Autobrains has raised just under $140 million so far and is taking an approach it believes will give it more market support due to its flexibility.

Many autonomous driving technologies (Mobileye is one example) are based on LIDAR sensors, and some companies (like Wayve) are building systems that combine experiences at a low cost using radar, smartphones and artificial intelligence. Autobrain takes a different approach, which can be described as device independent, using radar and LIDAR, but only if they are built into the OEM.

The company’s vision stems from over a decade of research and development. Basically, the startup originated from Cortica AI (founded by Rachelgause), which spent years creating AI-based imaging technology for applications in various cases (our first coverage of this actually developed. For advertising): Autobrain was originally developed as “Cartica AI to further realize the value of intellectual property as it relates to a very specific use of driving. The company says it has filed over 250 patents for its technology.

One of the main hurdles for unmanned AI is the inability of machine learning systems to account for edge cases, where decisions are made primarily based on labeled datasets fed into algorithms. “It’s a very expensive process that involves thousands of people, but it still runs into accuracy issues because you can’t cover all edge cases,” Rahelgauz said. So, in an unfortunate example, although the operator in the Uber accident involving a self-driving autopilot in Arizona was charged with the accident, the reason the car didn’t stop on its own was because it didn’t recognize the J-Walker.

As Rachelgauz describes it, AutoBrain doesn’t rely on labeled data and is designed to work “close to the human way of learning” by keeping the data random, allowing the platform to find matches, and then step through the lessons to keep what’s up to date. Teach (e.g. clothing the same color as the background) but ignore details that don’t match (e.g. the shape of the clouds). What is retained begins to form insight clusters that teach the self-managing platform to more accurately respond to related scenarios. For example, pedestrians can have over 100 different behavior classes developed in the AutoBrain system.

The platform is currently configured for two levels of unmanned vehicles. The first is for power-assisted systems aimed at improving driver safety, due to enter commercial service in 2023, which will raise the cost of a car to an average of $100. The second focuses on self-driving in levels 4 and 5 and is “working on it now” and will use any equipment built into cars to do the job. It is currently expected to cost “several thousand dollars” and production should start in 2024, but with the caveat that this may change depending on the market, customer appetite for investment, advances in technology, and of course that consumers actually want and use. . (A two-pronged approach, originally focused on AI-based driver assistance rather than autonomy scenarios, is also being used by other startups in the field: another machine learning startup called Enotel, for example, has recently also raised money.)

“I think it’s a process, not an immediate goal,” Rahelgauz said of the fully autonomous roadmap. “But if we can make a commitment by 2024, [we can so so understanding] It will take time to see how we can safely measure this. How it will be done is different for us. ,

“Autobrain technology is promising as we all strive to build an industry paradigm for AI machine learning and bridge the gap for fully autonomous driving,” said Tui Lin Pham, Deputy CEO of Winfast. “Autobrain has captured our attention by implementing untraceable AI software, as opposed to traditional software based on manually labeled data, to optimize unmanned vehicles for unprecedented real-time behavior. We hope that Autobrains will achieve this ambitious goal in the near future.”

Leave a Reply

Your email address will not be published.