Startup Aims AI System for Safety, with Potential for Self-Driving
Over the past 25 years or so, San Diego has become a hub for innovation in telematics, and in providing technologies and services that enable long-haul trucking companies and other fleet operators to monitor the location and operation of their vehicles.
Qualcomm (NASDAQ: QCOM) pioneered the use of satellite-based messaging and vehicle tracking technology in 1988 with its Omnitracs business (now an independent company based in Dallas, TX). Since then, San Diego startups DriveCam, Networkcar, and SmartDrive Systems have evolved into major providers of tracking, recording, and analytic services that help transportation companies analyze driver performance, manage risks, and improve their fleet efficiencies.
It’s turned into a big business, so much so that you might think innovation in vehicle telematics had run its course. The industry’s recent consolidations include the private equity firm GTCR’s $500 million buyout of DriveCam, now known as Lytx. In 2012, Verizon acquired Networkcar, (now operating as Verizon Networkfleet) as part of its $612 million buyout of Hughes Telematics.
But Sandeep Panya still sees plenty of room for innovation on the road ahead—especially as the development of self-driving cars accelerates.
As the president of Netradyne, a new San Diego startup, Pandya is focused on applying machine vision and machine learning technologies to the enormous amounts of data being generated by digital video cameras aboard thousands of trucks and other vehicles throughout the United States. The big idea at Netradyne is to use artificial intelligence technologies to help identify risky driver behaviors in real time—and to alert fleet managers accordingly.
“The insurance industry is looking for data on dynamic performance,” Pandya said. “It’s not enough to get an alert a day later or two days later. When you see a driver starting to struggle, a fleet manager can step in and call the driver and find out what’s going on.”
In a recent statement, Netradyne said it is incorporating AI modules made by Nvidia, the Santa Clara, CA-based chipmaker, enabling each dashboard unit to analyze video data generated simultaneously by four cameras aboard each vehicle (looking forward, backward, and to each side). Nvidia has emerged as a leader in machine learning tasks, to which its graphics processing chips are well-suited. By doing the machine learning aboard each vehicle, NetraDyne says only the most critical information is passed up to the cloud, where it can be relayed to fleet managers.
“We wanted to develop technology that solves a big data problem—and one of the basic problems in big data is [processing and analyzing] video,” Pandya said. “If you could take analytics into the device, you’d have a full-time reviewer in the cab all the time.”
Existing telematics technologies can alert fleet managers if a vehicle has been in an accident, but Pandya said Netradyne’s technology is intended to also issue real-time alerts for two other types of events—repeated hazardous driving, such as running three red lights in a row, and when a driver is showing signs of fatigue.
Although Netradyne was founded just 16 months ago, Pandya said the startup already has about 45 employees, including a core team of principals and engineers from Qualcomm and its former Omnitracs subsidiary.
Netradyne CEO Avneesh Agrawal, who held Qualcomm’s top job in India as senior vice president of technology and business in Bangalore, co-founded the company in September, 2015, with David Julian, a principal engineer deeply involved in Qualcomm’s Zeroth neuromorphic processor, a semiconductor designed to mimic the way a brain thinks. Pandya, a product manager overseeing Qualcomm’s Snapdragon and Gobi chipsets, joined a month later.
The startup raised $16 million last summer in a Series A round funded by the venture arm of Reliance Industries, a diversified conglomerate based in Mumbai, India.
While the company has initially focused its technology development on telematics for fleet management, Netradyne said at the time that Reliance Industries’ strategic investment would be used “to drive the commercialization of Netradyne’s intelligent Internet of Things (IoT) product roadmap… Crowdsourcing of visual IoT data continuously improves NetraDyne’s [sic] Deep Learning models, delivering insights and new levels of business intelligence to decision makers.”
In other words, long-haul trucking and fleet management is only the first stop on Netradyne’s technology roadmap. “We think there’s a lot of runway, and a lot of opportunities out there,” Pandya said.
In this respect, Netradyne’s strategy is borrowing a page from the Qualcomm playbook. Early revenue generated from its Omnitracs business enabled Qualcomm to advance its proprietary digital communications technology, and stage a demonstration for 50 wireless industry leaders in 1989. By 1992, Qualcomm was making CDMA-based wireless chips, mobile phones, and base stations.
Netradyne already has established partnerships with transportation companies to refine its technology, said Adam Kahn, vice president of fleet business. The startup has been generating data for a variety of fleets, from town car and package delivery services to long-haul trucking companies.
By developing AI to process enormous amounts of digital video technology in real time, Pandya explained that Netradyne will have an early lead in the race for autonomous vehicle technology with a system for generating and analyzing data in all driving conditions.
“When you train a neural net to detect certain things in video images, the images themselves are not interesting,” Pandya said. “But the training that goes into it [i.e. the machine learning] is interesting. That’s what we’re getting. We’re training our system to drive billions of miles.”