The Swedish Word for AV Tech

Veoneer, a new Tier 1 supplier with well-established roots, is moving rapidly into AI, says veteran research boss Ola Boström.

March 1, 2019

Lindsay Brooke

Veoneer VP of Research and Innovation Ola Boström earned his Ph.D in Theoretical Physics. In 2017, he was honored with the U.S. Government Award for Safety Engineering Excellence for his research on vehicle safety technology including advanced seat belt and air bag systems. He also is a pioneer in neck injury research. (Image: Veoneer)

The first automotive camera with built-in Deep Learning — considered to be a significant step forward in autonomous-vehicle sensor technology — is due to launch this year. It was developed and will be manufactured by a Tier 1 whose name is still seeking broader recognition in the industry.

“What is a...Veoneer?” asked an engineer waiting in the taxi line at this year’s CES, when a monorail train wrapped with the company’s logo flashed past above. Building brand cred among countless other spin-offs and tech start-ups who are clawing for visibility takes time, of course, even for the independent, publicly-traded Swedish company with 8,600 employees, formerly known as Autoliv’s Electronics segment.

“We are the largest pure player in ADAS — cameras, radars, ECUs, restraint control units,” claimed Ola Boström, Veoneer’s VP of Research and Innovation. “We’re now in lidar. Our existing product portfolio was a good starting point for supplying auto-makers around the world. More advanced features for Euro NCAP 2020 and [SAE] Levels 2 and above are in development and will be in the market soon. These will apply Deep Learning architectures,” he told SAE’s Autonomous Vehicle Engineering.

A branch of Machine Learning in artificial intelligence (AI) science, Deep Learning uses deep neural networks to conduct multi-object detection and classification for AVs. It is essential for operational edge-case decision-making and mapping. “AI is a growing chunk of our business and our hiring,” Boström said.

Deep Learning will be part of a new vision-system product launched by Veoneer this year. The technology is vital for multi-object detection and classification and operational edge-case decision-making. (Image: Veoneer)

It’s also woven into Veoneer’s strategic fabric. The company is a founding partner in Sweden’s “AI Innovation,” a new national incubator for AI. And together with Volvo Cars, it created a unique software company called Zenuity which Boström maintains will soon be a “powerhouse” in vehicle-sensor technologies and applications. “It was easy to recruit a great team of AI experts for Zenuity; we didn’t have to advertise,” he noted.

Veoneer is the only integrator of the Zenuity software. Volvo owns half, but has no exclusivity or first-to-market advantage.

Building trust in AVs

In addition to growing its internal expertise, Veoneer has recently made precisely-aimed tech partnerships. One, with Nvidia, resulted in development of a new supercomputer that Veoneer has yet to talk about in detail. Another is with Seeing Machines, whose FOVIO-based driver monitoring system is used by GM in the Cadillac CT6’s Super Cruise driver-assistance system. Boström’s engineers also are working with voice-recognition specialist Nuance.

Boström speaks enthusiastically about Veoneer’s collaborative work with both MIT’s AgeLab and Affectiva—the emotion-measurement company that grew out of the MIT Media Lab. Affectiva has developed software to recognize, via camera, human emotions based on facial cues or physiological responses.

“We take that technology and put it in the car,” he explained. “If you want to create trust, you have to have a collaboration, and the car needs to understand the persons in the car. That’s why we’re starting to use these technologies. How will it evolve? We’re trying to find out.”

“Building human trust in driver-assistance and autonomous systems is a primary challenge facing AV developers across the industry,” Boström asserted. “It starts with the systems in today’s SAE Level 2 ADAS-equipped vehicles,” he said.

“If a traffic speed sign is covered or missing, the car should tell the driver it can’t see the sign,” he offered. “According to the car’s map, the speed limit should be 10 mph because there’s a school close by. So, the car slows down now for you, but it can’t see the sign. The car admits it can’t see the traffic sign. For drivers, this is creating trust. You don’t have 100 percent reliable information about the speed limit, but there’s a communication to the operator. It’s not about 100% reliability; it’s about 100 percent transparency.”

SAE Level 3 currently is a big question mark in the AV development journey, Boström believes. “It might go away; we might not see any Level 3, so it goes back to a split between Level 2 and Levels 4 and 5.”

“In our testing, we see that people can either drive or ride. There’s no middle ground,” he said. “We’re working to make Level 2 driving simpler and simpler. Right now, in the industry we see a lot of confusing technology—the important thing with Level 2 is that the car is communicating to the driver/user.”

Boström said the 1.4 million traffic fatalities worldwide (in the face of rising numbers of road users) “is what Veoneer is attacking.” He joined Autoliv in 1995 and prior to the Veoneer spin-off was leading research activities in biomechanics, robotics, human-factor engineering and real-life traffic analysis.

“The first day I started at Autoliv, I was told by my boss at the time, ‘Don’t make the Research Dept. grow too large and centralized,’” as happened at some OEMs. For this reason, Boström is set on keeping his Veoneer team small. With tech centers in Japan, Korea, China, France, Germany, Sweden, and a large new facility in Southfield, Michigan, the unit also works with universities, with OEMs and other partners. However, he says the word ‘Research’ in his title, and what his team is doing, “is a bit of a misnomer.” A more accurate focus is integration, because “getting our new technologies into vehicles and onto the road, is what’s most important.”