Teraki Using Edge Processing to Meet Exploding Data Demands of AVs
Teraki, a Berlin-based data processing and artificial intelligence company working in the mobility sector, has had a busy fall.
In late September, Teraki raised a $3 million seed round from investors including Paladin Capital Group, GPS Ventures GmbH, and Deutsche Telekom’s hub:raum incubator. Earlier this month, the company announced its software would get its first commercial deployment in microcontrollers made by Infineon, a German semiconductor company.
Data processing with speed and accuracy will be key to getting autonomous vehicles (AVs) on the road, and though Teraki is a smaller, European player in the mobility industry, it’s working on solving an important part of the self-driving challenge with American automakers.
Daniel Richart, Teraki’s founder and CEO, says that as vehicles become more sophisticated and autonomous, they generate a staggering amount of data and analytics. Automakers, insurance companies, and even retailers see huge financial and safety opportunities in mining the data generated by cars, sensors, and drivers.
The problem, Richart says, is that chips used to process A.I. are expensive, and the computing demands of neural networks are onerous—to the point that they’re stifling the widespread expansion of automotive A.I. applications, he maintains. Also contributing to the problem, he says, are the limited processing power and bandwidth constraints of in-vehicle engine control units (ECUs), the high cost of data communication in car-to-cloud networks, and the time required to train machine learning components.
Richart says Teraki surmounts these data challenges with its signal-processing software, which the company says can “deliver a more than ten-fold increase in efficiency for automotive chip, communications, and learning performance in embedded environments.” Teraki uses edge processing— computing that is done at the source of the data as opposed to in the cloud. The company “downscales” cloud analytics models to fit and operate within a constrained environment like a car’s ECU. Teraki’s software can make things like threat detection, predictive maintenance, and indentifying driver behaviors more reliable, Richart says.
Self-driving cars are expected to generate 60,000 times as much data as the average smartphone today, he adds. “Cars have more and more sensors producing more and more data. To have autonomous vehicle functionality in the future, you can’t keep throwing power to the car’s computer.” .
Suppliers and car companies tell Teraki “they struggle with the rising cost of chips or big neural networks,” Richart says. So Teraki is working with a number of them—he wasn’t able to be specific due to nondisclosure agreements, but one is based in metro Detroit—to further refine its software. That testing will be carried out over the coming year.
The 30-person company, which was founded in 2015, hopes to eventually license its technology for non-automotive uses as well. “We do focus on the automotive market, but our technology could be adapted to any industry with the same challenges, like consumer robots or wearable devices,” says Geert-Jan van Nunen, Teraki’s chief commercial officer.