Here’s How Nvidia Plans To Ensure Self-Driving Car Safety
It’s all about redundancy.
Nvidia computers and software are working their way into self-driving cars from a number of companies. While those companies will ultimately have to ensure the vehicles they put on the road don't kill anyone, Nvidia is taking an active role in safety.
The company claims its latest self-driving car computing platform includes a number of redundancies that allow it to operate safely. On the hardware side, Nvidia packages different processors of different types so the system can keep functioning if one processor goes down. The same goes for software: The BlackBerry QNX-based operating system isolates individual applications so they don't interfere with each other.
The algorithms that control the system can juggle between the various processors as needed, Nvidia said. The company also leans on the triumvirate of sensor types—cameras, radar, and lidar—used in most self-driving cars to ensure safety. But while having three types of sensors adds redundancy, a car needs all three to get a full picture of the environment. Cameras, radar, and lidar all have weaknesses that the three-sensor arrangement compensates for.
Beyond the design of hardware and software, another important factor in self-driving car safety is testing, Nvidia noted. Like Waymo, Nvidia relies heavily on simulations to expose self-driving car systems to specific scenarios they might not encounter in the real world. Tests on public roads are important, but developers can't guarantee that self-driving cars will encounter everything they want to test.
Enough companies are developing self-driving cars that the mere existence of these vehicles is no longer remarkable. From Ford to Waymo, the number of autonomous prototypes on public roads proves that, to some extent at least, computers can drive cars. Now companies must prove that this technology can work safely and reliably, and not just during carefully-choreographed demonstrations.