Photo credit: arstechnica.com

Volvo Enhances Autonomous Driving Technology Through In-House Development and AI

Volvo is pioneering advancements in autonomous driving technology (ADT) by developing its software in-house, a strategic shift that allows for greater agility in the innovation process. Erik Coelingh, the Vice President of Product at Zenseact, a subsidiary specializing in advanced driver-assistance systems (ADAS), discussed the benefits of this approach in a recent statement.

“One major advantage is that we now handle development internally,” Coelingh explained. “This minimizes reliance on external suppliers and eliminates the lengthy back-and-forths that often delay projects. If we encounter any issues, we can address them within a day.”

This rapid development cycle has fundamentally transformed Volvo’s innovation speed. “We’re iterating much more quickly,” Coelingh noted. “As Bakkenes pointed out, we conduct daily tests with our new software, allowing us to advance safety measures and significantly aim for reduced accident rates at an unprecedented pace.”

A key benefit of transitioning to Software-Defined Vehicles (SDVs) is the ease of simulation. The company has established one of Europe’s largest data centers to facilitate complex virtual simulations of their vehicles. These simulations enable Volvo to directly test the integrated software stack under various conditions.

In its pursuit of rapid development, Volvo has also harnessed the power of artificial intelligence (AI). However, the challenge remains: how can such a safety-focused organization ensure the reliability of AI-generated outputs?

Coelingh elaborated on this by introducing the concept of Gaussian splitting, a technique that allows Volvo to transform a single real-world traffic scenario into thousands of test cases. “With this technology, we can expand a specific scenario into tens of thousands, manipulate those variations, and then thoroughly test our software against them,” he explained.

Furthermore, the use of advanced simulation environments, such as Unreal Engine, has been a common practice among autonomous vehicle developers for visual scenarios. However, Coelingh emphasized Volvo’s approach: “While visual simulations are beneficial for camera data, we also utilize lidar, radar, and camera data to reconstruct scenes using neural networks. This enables us to conduct closed-loop simulations effectively.”

By leveraging these methods, Volvo can rapidly assess its software performance across a vast array of realistic driving scenarios, further enhancing their commitment to safety and technological advancement in the automotive sector.

Source
arstechnica.com