Volvo EX90 Uses AI, NVIDIA SoC to Make Automated Driving Safer

April 26, 2024

Sebastian Blanco

As head of software engineering at Volvo Cars, Alwin Bakkenes is involved not just with all of the software and electronics in Volvo’s vehicles but also the automaker’s automotive cloud, the data center that trains Volvo’s algorithms, the connectivity pipeline and software updates as well as interactions with Volvo’s autonomous driving software development subsidiary Zenseact and HaleyTek, a joint venture with ECARX to develop Android-based infotainment systems for Volvo and Polestar. This growing digital footprint gives Volvo an array of tools to improve its future vehicles, something Bakkenes made clear when speaking with SAE Media at the 2024 NVIDIA GTC event in San Jose in March.

Volvo started working with NVIDIA around eight years ago and first used the NVIDIA DRIVE Orin system-on-a-chip (SoC) technology in the updated XC90 SUV, introduced in 2022. In 2023, Volvo built a new 22,000 sq m (236,806 sq ft) software testing center in Sweden at a cost of around SEK 300 million (U.S. $28.4 million).

It’s in Volvo’s new EX90, though, where the NVIDIA effect can be best seen. The all-electric SUV is equipped with NVIDIA DRIVE technology and a core compute architecture capable of over 280 TOPS. The architecture consists of two core computers that allow Volvo to make the EV “an interface-first, software-defined product where we build and drive a lot of the behaviors of the actual vehicle,” Bakkenes said. The architecture also runs the car’s ADAS and autonomous driving stack, which Volvo developed in-house “based on our heritage of having done years of safety research and pioneering technology development in terms of active safety.”

NVIDIA inside, and also outside

Volvo uses NVIDIA components not just in the car but also in its data center, which is heavily DGX-based, and on desktops. Faster compute speeds, and more sensors in Volvo vehicles allow the developers to work at a pace that would have been impossible before.

“We’ve always innovated but, typically, these innovations took years to go to market,” Bakkenes said. “With the technology that we’re introducing [in the EX90], we can actually improve the systems in a matter of weeks. And from a pure developer perspective, we do it in a matter of days. That ability to iterate and learn from a large fleet of cars, in terms of how the systems behave, enables us to be really fast and deliver better cars every day.”

The entire auto industry now has access to AI and more powerful computers. Even though everyone can improve their development speeds this way, Bakkenes said Volvo isn’t worried about other OEMs catching up to Volvo’s real and perceived lead in automotive safety because Volvo comes at the autonomy problem from a specific perspective.

“If anything, [AI] will enable us to actually leap frog even more,” he said. “Our purpose is to build the safest products out there. Collision avoidance is super important and this is where we start. We focus on autonomy, as well, but autonomy for the purpose of creating safer products. We have a slightly different heritage perspective on autonomy than many others, and that gives us an edge.”

One way Volvo expects to use AI to make its vehicles safer is by harnessing the increased data that the EX90 and future Volvos will provide, including details on actual events and near misses. The EX90, for example, is equipped with exterior radars, exterior cameras, interior cameras, a driver monitoring camera, an interior radar, and lidar on the roof. Some of this information collected can, with user consent, be anonymized and sent to Volvo for further analysis in ways that simply were not possible before. When an incident occurs, Volvo can get information about objects around the car and can see certain things that the sensors saw, which then allows Volvo to recreate the scenario in digital form.

“Instead of understanding what happened, we can actually send the actual sensor data, which we can use for actual retraining of new scenarios, so we can be even better at improving the system over time,” Bakkenes said. “We can find corner cases of scenarios that we do not support and we can actually use the technology to build these scenarios out, we can simulate them, we can retrain algorithms to cover them to make the performance of the systems better.”

These enhancements can then be deployed back to cars on the road and be used in future vehicles.

“It’s a mindset thing,” he said. “We’re focused on making our products safer. So when we leverage data, we do it with the intent to actually understand how can we make it better, how can we make it safer.

“Starting in the EX90 with our core compute architecture, we can actually start to use some of the data that we get back to recreate the scenarios and actually train algorithms on real-life scenarios. A fleet of actual products can capture reality much better than engineers driving in pre-production vehicles, so it gives us real insights about real-life safety and how to make the products better.”

---