Enabling an AV driver’s license test
In the Wild, Wild West of the current AV (autonomous vehicle) industry, many claims are made by the industry’s OEMs, suppliers, and technology providers regarding vehicle and system performance capabilities. Much effort is expended verifying those claims and making purchase decisions, whether that be by industry specifiers or the end consumer. However, more help might be on its way—in the form of simulation.
At its GPU Technology Conference in March, Nvidia announced that its Drive Constellation autonomous vehicle simulation platform is now available after being discussed at last year’s event. The cloud-based platform enables millions of miles to be driven virtually across a range of scenarios, with claims of greater efficiency, cost-effectiveness, and safety than what is possible to achieve in real-world physical testing.
The data-center solution comprised of two side-by-side servers, one using Nvidia GPUs (graphical processing units) running Drive Sim software to generate sensor output from a virtual car driving in a virtual world, the other containing the Drive AGX Pegasus AI car computer, which processes the simulated sensor data. The driving decisions from Drive Constellation Vehicle are fed back into Drive Constellation Simulator, enabling bit-accurate, timing-accurate hardware-in-the-loop testing.
As the industry has discovered, simulation is a key component of any AV testing program, enabling many more scenarios and iterations of technology development to accumulate billions of virtual miles and speeding safety validation. It is also enabling development of third-party autonomous vehicle standards, as evidenced by how TÜV SÜD is working with the Nvidia platform to formulate its self-driving validation standards.
The German testing and certification organization is building “a comprehensive suite of tests that ultimately could lead to some type of autonomous vehicle driver's license to ensure that an AV can handle everything it needs to handle before it goes out on the road,” said Danny Shapiro, Senior Director of Automotive at Nvidia, in a media briefing just before the GTC event.
Last year, TÜV SÜD announced a cooperation with Nvidia and AVL GmbH to develop the AV certification process, the goal which is to validate and establish simulation as a method and approval tool for future homologation. Its experts estimate that there are up to 100 million situations that should be evaluated to determine self-driving capability. The time and costs to perform traditional on-road tests to evaluate these critical scenarios is not practical, so the virtual world will have to take a major role in augmenting established physical methods. Performing simulation testing at the necessary scale requires enormous computing capacity, which Nvidia’s technology provides.
In a GTC 2019 presentation, Dr. Houssem Abdellatif, Head of Automated Driving and Driver Assistance Systems at TÜV SÜD Mobility, described what’s needed to validate fully automated vehicles and technologies, including the critical scenarios, the connection between physical and virtual testing, and functional safety—with an eye to international regulations for homologation.
Together with the DFKI, the German Research Center for Artificial Intelligence, his organization is developing a “TÜV for algorithms” in collaboration with the testing-ground operator in German state of Baden-Württemberg and the Karlsruhe Transport Association. It is also working with CETRAN (Center of Excellence for Testing & Research of AVs at Nanyang Technology University in Singapore) on the world’s first standard for authorization of fully automated vehicles in Singapore.
The ability to drive billions of miles in virtual scenarios will be critical to bringing safe, automated driving solutions to market. Simulation will ensure that automated driving functions are executed accurately, reliably, and safely—one of the greatest challenges in approval of autonomous vehicles. The TÜV SÜD plan is to apply the findings as early as the vehicle development phase in the future, ensuring that driving functions will demonstrably comply with regulatory requirements right from the outset, said Abdellatif.
The good news is that the simulation wheels are in motion for an automated vehicle driver’s license. The effort would enable manufacturers and regulators to validate that self-driving technology will behave as intended. It would not only make transportation safer, sooner, but also enable greater transparency as well as confidence in and adoption of life-saving automonous vehicle technology.