Autobrains Founder and CEO Igal Raichelgauz is on a mission to transform the automotive industry by putting Artificial Intelligence (AI) in the driver’s seat. But this vision is not simply about integrating AI into the driving experience; Raichelgauz and his team at Autobrains recently unveiled Liquid AI, a paradigm-shifting, self-learning AI technology that could help put fully autonomous vehicles (AVs) on the road in less than a decade.“A mass-produced AV is possible by 2035,” explains Raichelgauz. “It still will not be 90 percent of the market, but it will start to be significant numbers.”A potentially bold prediction—possibly—given that consumers have been hearing prognostications for fully autonomous vehicles since 2016, when CEO Elon Musk predicted that a Tesla would be able to drive from Los Angeles to New York City without human hands on the wheel by the end of 2017.So, what’s the hold up? For Raichelgauz and the Autobrains team, it’s in how the technology learns. A leader and innovator in the field of artificial intelligence, and, more specifically, in the development of AI technologies that enable AI to learn on its own, Raichelgauz believes a self-learning approach is more likely to close the AI gaps that are preventing more widespread use in the automotive sector. An award-winning innovator with dozens of patents, Raichelgauz paused his Ph.D. research in the field of cortical neural networks to found Cortica, one of Israel’s leading AI companies, and, in 2019, to found Autobrains. “Automotive is a very demanding space for AI,” says Raichelgauz. “It is not about generating text or images, or searching images, it’s about working within a very complex, real-world environment with a lot of role players, like pedestrians, drivers, cars, and road conditions. This combination of factors is extremely challenging for AI.” It is exactly this complexity that inspired Raichelgauz and the Autobrains team to focus on creating disruptive AI technology that addresses autonomous driving challenges. The result is Liquid AI, which combines a signature-based self-learning AI approach with a modular and adaptive architecture of specialized, scenario-based end-to-end skills. Liquid AI overcomes many of the challenges faced by conventional AI—including training AI to be prepared for a seemingly infinite variety of unexpected driving scenarios; the escalating costs and power consumption involved with adding data, layers and labeling to existing systems; and the interplay of perception and decision functions that better ensures effective decision-making. “We are convinced that this is the only technology today that can really solve this challenging problem with the longtail of edge cases and the amount of compute and ultra-accuracy needed to perform,” explains Raichelgauz. “Our approach is very different. We don’t use leveraged training, and, as such, we don’t generate a generic system that works well on average. We can scale to a longtail of edge cases without investing exponential resources.” In other words, Liquid AI enables cars to learn, collaborate and interact with the world much like humans, but without human input. The technology’s unsupervised learning capabilities enable a more detailed, comprehensive and precise interpretation of the car’s surroundings, in real time, resulting in better performance, with less cost and energy consumption. “The vision came through seeing the gap between the most advanced AI systems and the capability of biological systems of the human brain, or, more generally, the mammal brain,” explains Raichelgauz. “The inspiration was to understand not necessarily the specific biological details, but the computational principals and biological systems that we can learn from, and then replicate and simulate them to next-generation AI systems.” Autobrains’ innovative technology has also caught the eye of a Chinese Electric Vehicle manufacturer, resulting in a partnership that will see Autobrain’s Liquid AI deployed in a new line of vehicles, with production slated to begin Q4 2024. The collaboration has the potential to redefine industry standards and accelerate the adoption of advanced driver-assistance systems (ADAS) across the automotive industry. “This is a very important partnership,” says Raichelgauz. “Once we have those cars on the road, generating millions and millions of miles of data, it will prove that this technology—with still very limited resources, sensors and compute—can solve the unsolved problems in the industry related to edge cases.” Further, Raichelgauz believes that the insights derived from this partnership also will show other industry players that self-learning AI is the path to achieving Level 3 and Level 4 technologies. For those new to the self-driving nomenclature, SAE international, formerly known as the Society of Automotive Engineers, has identified six levels of self-driving automation. Levels 0 (no automation) to Level 2 mean that a person is still driving and is legally responsible for the vehicle, with the self-driving features available for driver support. At Level 3, the driver is not actively driving, but still could intervene in specific situations. It is at Level 4, or High Driving Automation, that vehicles operate in self-driving mode. Waymo is a good example of this, offering Level 4 robo taxi service in specific areas within Phoenix, Los Angeles, San Francisco, and Austin, Texas. Level 5 is envisioned to be fully autonomous and projected to not include human override features such as steering wheels or brake pedals. They also will be free of the geofencing currently used by today’s robo taxi services, able to travel on any road and do anything that an experienced driver could do. For Raichelgauz and the Autobrains team, Liquid AI will be the technology that propels ADAS and AV technologies from Level 2 to Level 5. By circumnavigating the issues preventing traditional AI-based systems from making a fluid progression from ADAS to fully autonomous, Liquid AI is posed to not just get into the race—but take a pole position. It also has the possibility to change the trajectory of autonomous driving from a long-term possibility to a more near-term reality, and integrate precision, insight, and affordability into the experience. The transition from ADAS to AV is likely to be a gradual path to progress, both for the technology as well as for mass adoption, Raichelgauz believes. “We will start to see more and more partial autonomy, meaning the car will be full autonomous, but only under certain conditions and scenarios, and that coverage of scenarios will gradually scale,” he explains. “Actually, our technology is all about this transition. We call it driving skills, and we will start with a certain amount of driving skills, but there will be some scenarios we will not be supporting, and the car and the AI will know what it doesn’t support so that the driver will be able to take control ahead of time.” Over time, as Raichelgauz and the Autobrains team collect more data, the Liquid AI technology will develop new skills, which will be uploaded into existing infrastructures and gradually increase the vehicle’s coverage. It will be within this timeframe that drivers will get comfortable with how the technology works. “As drivers use the features every day and see no challenges, then the transition becomes smooth,” says Raichelgauz. “You just become used to the technology and think there’s no magic, it just becomes how things should work.” Which could give drivers a lot more time back in their day. “It definitely makes sense to let AI do more of these types of tasks, which allows people to do more interesting things with their time,” Raichelgauz concludes.
Copyright © 2024 California Business Journal. All Rights Reserved.