In a new report, Nvidia joined a growing set of companies to release a voluntary safety self-assessment under guidelines set by NHTSA.
The chip maker's report is the first from a supplier of automated drive technology hardware and development infrastructure, rather than a company developing a self-driving car. The report focuses on safe development practices and the ways in which Nvidia's hardware reinforces the safe operation of automated drive systems.
"We are different in that we're not putting a fleet of vehicles on the road for others," said Danny Shapiro, senior director of automotive at Nvidia. "We're a layer down. We're the fundamental technology that other companies build on. But in everything we're doing, from collecting data to training to simulation to building the hardware that goes in the car, we have a maniacal focus on ensuring that we develop is as safe as it possibly can be."
Nvidia's report highlights the functional safety measures built into its in-car hardware and software and its development best practices and the role that its ecosystem plays in ensuring that autonomous vehicles are as safe as possible. Because Nvidia supplies virtual safety validation in simulation, its safety report describes the role that these systems play in establishing safe algorithms before cars start testing on public roads.
"Simulation allows us to ensure the safety of the system, prove the technology and build trust with the public," Shapiro said. "We're going to be able to catch a lot of things we wouldn't be able to catch otherwise, or which would take forever to see in the real world. We're not racing to put something out there as fast as we can."
The recent fatal crash of an Uber autonomous test vehicle in Tempe, Ariz., was "a turning point in terms of safety awareness for the industry" said Shapiro. In addition to demonstrating the need for more simulation testing before the start of public road testing, the Tempe crash is also creating more demand for sheer computing power in autonomous cars, Shapiro said.
"Diversity and redundancy of sensors and many different data types, processors for that data and different kinds of algorithms are what gives you safety improvements," he said. "So you really need more computing power."
Even driver assistance systems are starting to use Nvidia chips once thought of as being autonomous drive-grade, Shapiro said, pointing to Volvo's decision to use Nvidia's Xavier autonomous drive board for its next-generation advanced driver assistance systems, which debut in 2020. Recent testing by Consumer Reports, IIHS and others has shown that current lane-keeping and adaptive cruise control systems struggle with many real-world conditions, showing the need for ever more-powerful hardware.
"There were a lot of failures," Shapiro said. "Cars were not performing well. Volvo and others are looking at that and thinking even to deliver those kinds of features you need more sensors, you need more compute, you need more AI."
