- Volvo’s partnership with Nvidia goes beyond chips in cars
- Gaussian splatting creates high-fidelity 3D scenes
- The technique can recreate “edge cases” to train models faster
Volvo is using a new AI technique called ‘Gaussian splatting’ to train its vehicles and accelerate its goal of zero collisions on the roads – and it’s all thanks to its recently expanded partnership with Nvidia.
Last month we reported that the upcoming Volvo ES90 will be the most powerful car it has ever created in terms of core computing capacity, due to it packing a dual Nvidia AGX Orin configuration.
Now, the company has revealed how this sort of supercomputing is also helping it to more quickly train its Advanced Driver Assistance Systems.
Volvo claims that it can now synthesize incident data collected by the advanced sensors in its latest vehicles, such as emergency braking, sharp steering or manual intervention.
This then allows the company to reconstruct and explore them in new ways to better understand how incidents can be avoided.
The novel method is dubbed Gaussian splatting and it allows the company’s software to produce realistic, high-fidelity 3D scenes and subjects from real-world visuals.
Once these scenes have been created, Volvo’s engineers can manipulate them to generate a number outcomes. The video clip examples the Swedish marque provides are freakishly realistic.
It’s akin to a human learning how to skateboard by playing Tony Hawk’s Pro Skater for hours and hours on end.
“We can select one of the rare edge cases and explode it into thousands of new variations of the scenario to train and validate our models against,” Alwin Bakkenes, Head of Global Software Engineering at Volvo Cars, explains.
Bakkenes says this has the potential to unlock a scale that Volvo has never had before and even to catch edge cases before they happen in the real world.
Now the computers are training the computers
Gaussian splatting is a relatively new 3D rendering technique that doesn’t rely on neural networks, unlike more complex methods such as Neural Radiance Field (NeRFS).
This allows for incredibly complex 3D scenes to be created in real time. The technique is currently being explored in multiple industries, from gaming to interactive app development.
Volvo’s use of advanced Lidar, sensor and high-definition camera technology, as first showcased in the EX90, collects reams of data that can then be reproduced in a manipulatable 3D model, which allows its engineers to then train the vehicle’s AI to perform better in the real world.
There was some disappointment when the EX90 launched, seeing as its Lidar technology would remain offline for consumer use, effectively banished to merely collecting data until Volvo’s compute power was at a level where the company was happy to introduce ADAS systems that rely on the sensor suite and software stack.
Thankfully, its recently announced partnership with Nvidia will help the Swedish marque, which is synonymous with road safety, to realize its vision of zero collisions and driver assistance systems that actually help, rather than simply nag.
What’s more, the company has also stated that early EX90 models will be updated with the dual Nvidia AGX Orin System on a Chip set-up, so they too can make the most of the latest developments in autonomous driving and ADAS systems.
