TLDR: Volvo is revolutionizing its vehicle safety software development by leveraging generative AI and advanced virtual worlds. Through a technique called ‘Gaussian splatting,’ real-world incident data is transformed into lifelike 3D simulations, allowing the company to rapidly test and refine advanced driver assistance systems (ADAS) against millions of complex and rare ‘edge cases.’ This innovative approach, developed with in-house AI company Zenseact and supported by NVIDIA technology, significantly reduces testing time from months to days, aiming to create even safer cars.
Volvo Cars is pioneering a transformative approach to vehicle safety software development, integrating generative artificial intelligence and highly realistic virtual environments to enhance the robustness and reliability of its advanced driver assistance systems (ADAS). This strategic shift marks a significant evolution from traditional physical and early virtual testing methods, enabling the Swedish automaker to simulate an unprecedented array of complex driving scenarios.
At the heart of this innovation is an advanced computational technique known as ‘Gaussian splatting.’ This technology allows Volvo to convert real-world visual data into lifelike, natural-looking 3D scenes and subjects. Incident data, such as emergency braking maneuvers, sharp steering inputs, or manual interventions, collected by the sophisticated sensors in Volvo’s new vehicles, can now be synthesized, reconstructed, and thoroughly explored within these virtual realms. This capability provides invaluable insights into how accidents might be avoided in the future.
The virtual environments offer unparalleled flexibility. They can be manipulated to add or remove road users, alter traffic behavior, or introduce obstacles, exposing Volvo’s safety software to a vast spectrum of traffic situations at a speed and scale previously unattainable. This is particularly crucial for testing ‘edge cases’ – rare yet potentially dangerous scenarios that are difficult and time-consuming to replicate in real-world testing. Volvo states that this method can reduce the time required to expose software to such critical edge cases from months to mere days.
Alwin Bakkenes, Head of Global Software Engineering at Volvo Cars, emphasized the profound impact of this technology, stating, ‘We already have millions of data points of moments that never happened that we use to develop our software. Thanks to Gaussian splatting we can select one of the rare corner cases and explode it into thousands of new variations of the scenario to train and validate our models against. This has the potential to unlock a scale that we’ve never previously had before and even to catch edge cases before they happen in the real world.’
These virtual environments serve as a vital complement to real-world testing, providing a safe, scalable, and cost-effective platform for software training, development, and validation. The sheer number of potential scenarios – from animals darting onto the road, objects falling from trucks, or vehicles driving the wrong way, to extreme weather conditions like flooded roads or dust tornados – is virtually infinite. Capturing or manually coding all these possibilities would be an undertaking of hundreds of thousands of years, underscoring the necessity of AI-driven simulation.
The development of these virtual environments is spearheaded by Zenseact, Volvo’s in-house AI and software company, founded approximately five years ago. This project is also part of a PhD program sponsored by the Wallenberg AI, Autonomous Systems and Software Program (WASP), exploring the integration of neural rendering techniques into future safety initiatives.
Also Read:
- Waymo Integrates Generative AI While Upholding LiDAR and Radar as Core to Self-Driving Safety
- Generative AI in Automotive Sector Projected to Reach $2.6 Billion by 2032 Amidst Rapid Innovation
Furthermore, Volvo Cars’ ability to explore technologies like Gaussian splatting is bolstered by its expanded relationship with NVIDIA. The new generation of fully electric Volvo cars, built on NVIDIA accelerated compute, collects extensive sensor data to create a comprehensive understanding of the vehicle’s surroundings. An AI supercomputing platform, powered by NVIDIA DGX systems, then contextualizes this data, unearths new insights, and trains future safety models, significantly improving and accelerating AI development. This supercomputing platform is part of a recent joint investment by Volvo Cars and Zenseact to establish one of the largest data centers in the Nordics. The exploration of Gaussian splatting and generative AI was also a key part of Volvo’s presentation at the NVIDIA GTC conference.
