NVIDIA publishes its self-driving safety report
NVIDIA published its Self-Driving Safety Report, which it submitted to National Highway Traffic Safety Administration (NHTSA) as part of the safety self-assessment laid out in the agency’s voluntary guidance, "Automated Driving Systems: A Vision for Safety." The report is a look at how NVIDIA develops its autonomous driving technology.
The document presents a view into NVIDIA’s development processes, showing how the company is harnessing graphics processing units (GPUs) to create self-driving systems. Achieving the highest levels of computing allows the incorporation of diversity and redundancy throughout the process, including all aspects, from sensors to processors to algorithms. This ensures multiple lines of defense should a failure occur.
According to NVIDIA Senior Director of Automotive Danny Shapiro, “As a solutions provider to the vast majority of vehicle makers, suppliers, sensor makers, startups, and mapping companies in the autonomous driving space, NVIDIA makes safety our first priority. And we have integrated it into every step of the development process.” He added, “We believe safety is at the heart of the transition to autonomy.”
The report details how compute performance translates to safety at all stages, from initial data collection to public road testing. Featured in the report are the four pillars of safe autonomous driving, which are discussed briefly here.
Pillar 1 is an artificial intelligence design and implementation platform. A safe artificial intelligence (AI) driver requires a compute platform that spans the entire range of autonomous driving, from assisted highway driving to robotaxis. It must combine deep learning, sensor fusion, and surround vision to enable the car to make split-second decisions based on massive amounts of data.
Pillar 2 encompasses the development infrastructure that supports deep learning. A single test vehicle can generate petabytes of data annually. Capturing, managing, and processing this massive amount of data, for not just one car but a fleet, requires an entirely new computing architecture and infrastructure.
Pillar 3 is the data center solution for robust simulation and testing. The ability to test in a realistic simulation environment is essential to providing safe self-driving vehicles. By coupling actual road miles with simulated miles in a high-performance data center solution, manufacturers can comprehensively test and validate their technology.
Pillar 4 consists of a best-in-class, pervasive safety program. Self-driving technology development must follow a pervasive safety methodology that emphasizes diversity and redundancy in the design, validation, verification, and lifetime support of the entire autonomous system. These programs should follow recommendations from federal and international agencies such as the NHTSA, International Organization for Standardization, and the global New Car Assessment Program.
This safety program includes a comprehensive evaluation process before the safety drivers validate NVIDIA technology on public roads. Drivers and co-pilots must complete rigorous training before operating vehicles, which are continuously tested for hardware and software readiness before drives.
Shapiro noted that in addition to these four pillars, significant research and development, as well as industry-wide collaboration, is essential to safely deploying autonomous vehicles. He said, “NVIDIA continues to work with the wide and diverse ecosystem as well as regulators to share knowledge and formulate standards for this budding technology.”