Helm.ai raises $ 13 million for its unsupervised learning approach to driverless auto-AI


Four years ago, mathematician Vlad Voroninski saw the opportunity to solve some of the bottlenecks in the development of autonomous vehicle technology thanks to breakthroughs in deep learning.

Well, Helm.ai, The startup, which he founded together with Tudor Achim in 2016, comes from the hidden with the announcement that it has raised $ 13 million in a starting round that includes investments by A. Capital Ventures, Amplo, Binnacle Partners and Sound Ventures. Fontinalis Partners and SV Angel. More than a dozen angel investors also participated, including the founder of Berggruen Holdings, Nicolas Berggruen, Quora Co-founders Charlie Cheever and Adam D’Angelo, professional NBA player Kevin Durant, General David Petraeus, co-founder and CEO of Matician, Navneet Dalal, managing partner of Quiet Capital, Lee Linden, and co-founder of Robinhood, Vladimir Tenev , amongst other things.

Helm.ai will use the $ 13 million seed capital for advanced engineering and research and development, employ more people, and sign and contract with customers.

Helm.ai focuses exclusively on the software. The computing platform or sensors that are also required in a self-driving vehicle are not created. Instead, it’s agnostic for these variables. Basically, Helm.ai is developing software that tries to understand sensor data as well as a human would do to drive, said Voroninski.

This goal is no different from other companies. Helm.ai’s approach to software is remarkable. Autonomous vehicle developers often rely on a combination of simulation and testing on the road, as well as tons of data sets that have been commented on by people to train and improve the so-called “brain” of the self-driving vehicle.

Helm.ai says it has developed software that can skip these steps, which speeds up the timeline and lowers costs. The startup uses an unsupervised learning approach to develop software that can be used to train neural networks without the need for extensive fleet data, simulations or annotations.

“There is this very long tail and an endless sea of ​​corner cases that have to be run through when developing AI software for autonomous vehicles. Voroninski explained. “What really matters is the efficiency unit, how much it costs to solve a particular corner case, and how quickly can you do this? And that’s the part where we really innovated. “

Voroninski became interested in autonomous driving at UCLA for the first time, where he learned something about the technology from his study advisor who had participated in the DARPA Grand Challenge, a driverless car competition in the United States funded by the Defense Advanced Research Projects Agency learned. And while Voroninski devoted his attention to applied math for the next decade – he did his PhD in mathematics at UC Berkeley and then switched to the faculty of the MIT math department – he knew that he would eventually return to autonomous vehicles.

By 2016, Voroninski said, breakthroughs in deep learning created opportunities to get started. Voroninski left MIT and Sift Security. A cybersecurity startup that was later taken over by Netskope to launch Helm.ai with Achim in November 2016.

“We identified a number of key challenges that we believe traditional approaches have not addressed,” said Voroninski. “We built some prototypes early on that made us think we could actually do it.”

Helm.ai is still a small team of around 15 people. The business goal is to license the software for two use cases: Advanced level 2 driver assistance systems (and a more recent term called level 2+) in passenger cars and level 4 autonomous vehicle fleets.

Helm.ai has customers, some of whom have gone beyond the pilot phase, said Voroninski, adding that he could not name them.