Specialists used the recent Nvidia AI Summit to highlight a pressing need for standardised regulations for the autonomous vehicle sector, an area they acknowledged is undergoing a period of rapid technology-driven transformation.
During the event in Washington, DC last week, Nvidia VP of automotive Danny Shapiro (pictured, left) highlighted the high number of crashes, injuries and fatalities on the world’s roadways, with human error a primary cause.
“Improving safety on our roads is critical,” Shapiro stated, noting Nvidia’s two-decade-long collaboration with the automotive industry in developing advanced driver assistance systems and fully autonomous driving technology.
Shapiro explained Nvidia employs specialised computers to train the AI used in its vehicle systems, along with conducting simulation and testing, with a further machine fitted into vehicles to process sensor data in real-time.
These systems enable continuous development cycles, enhancing the performance and safety of AV software and, in turn, the vehicles themselves.
Mark Rosekind, a former administrator of the US government’s National Highway Traffic Safety Administration (pictured, right), drew a connection between safety and current approaches to regulation, citing his domestic market as an example.
The expert explained the bifurcated regulatory framework in the US could hamper development of comprehensive safety standards for AVs: the national government currently focuses on vehicle specifications while state-level equivalents handle driver training, insurance and licensing.
Technology shift
There was also attention on a need for regulations to be adaptable to keep pace with the changing technology landscape.
Marco Pavone, director of AV research at Nvidia (pictured, centre), noted generative AI and neural rendering technologies are enabling development of new tools which provide researchers and developers with the capability to rethink and enhance the AV development process.
These technologies are changing the way AVs are designed, tested and deployed, contributing to safer and more efficient autonomous driving solutions, Pavone asserted.
Attention also turned to how recent advances in simulation technologies are paving the way for more comprehensive testing of AVs, enabling the generation of complex scenarios to stress test vehicles for safety purposes.
Pavone noted employing foundational models including vision language enables developers to build more resilient autonomy software.
An example of the advances being made came in the form of a partnership announced during the Nvidia AI Summit between US-government backed non-profit MITRE and Mcity, a testing hub operated by the University of Michigan, which targets development of a virtual and physical AV validation platform.
MITRE will employ Mcity’s simulation tools and a digital twin to provide a platform offering physically based sensor simulation enabled by Nvidia Omniverse Cloud Sensor RTX APIs.
The approach is expected to enable developers to perform exhaustive testing in a simulated world, ensuring AVs are safely validated before real-world deployment.
Rosekind believes the MITRE announcement offers an opportunity to have a trusted source create an independent, neutral setting for safety assurance testing, something of a continuation of the overall focus on the need for consistent regulations.
He noted MITRE has a proven track-record in various sectors, especially aviation, which bolsters the credibility of the AV initiative.
There was a consensus simulation in AV testing provides benefits, with the speakers pointing to the ability to test very dangerous conditions in a repeatable manner.
Pavone underscored such capability is crucial for simulating different cases at scale, while Shapiro noted the repeatability and controllability of simulation are its greatest strengths.
In a simulated environment, developers can manipulate variables including the weather or time of day, and inject various hazards, a flexibility enabling them to run scenarios multiple times to guide software refinement and capability.
Comments