Nvidia has calculated an expected tracking range for cars, delivery vans, semitrailers, bulldozers, motorcycles, bicycles, and pedestrians walking beside and across a road... even dogs and cats. So it can predict future motion. We saw this technology running in the Nvidia booth a year ago at the 2016 Consumer Electronics Show (“CES”) in Las Vegas. We walked in front of a car at a crosswalk, and we could see our speed and trajectory uploaded into a “game” that a smart car was playing. So just by integrating fixed objects like streets, curbs, and stop signs with the moving-objects data, a self-driving car can reasonably navigate a cityscape. Nvidia is taking lessons learned in video games and applying them to real life... Obviously, we were impressed. But Nvidia’s latest technology is going a step further. We believe its technology will become the future of the Internet and mobility. Technically this process is called “deep learning,” but it’s easier to think of it as “self learning”... The next step for a car is to learn to drive from “scratch”... without a programmed database of every curb, stop sign, and left- turn lane stored in a memory bank. All it uses is a map... For months, Nvidia’s car “watched” data of what real-life drivers did in reaction to different scenarios. Then Nvidia took the car out to a closed course and let it start hitting cones. After a while, it started hitting fewer cones... and then no cones. THIS CAR LEARNS LIKE A STUDENT DRIVER
From there, the research team let it drive around cemeteries with no street names and no curbs. Then the car went into suburban streets. And finally, it started driving along the New Jersey Turnpike. This is also the logic that will help a car negotiate snow, ice, and slush as proficiently as any human driver. That’s because the car can do more than just learn from its own experiences... It can also upload data from other cars. Now, think back to our example of liability for a self-driving car. No command lines exist that order a car to leap the curb to avoid a head-on collision, but which regrettably wipe out a bus stop. Because there’s no command code, there’s much less liability. In summary, self-thinking cars can out- think a fixed-object program that’s based on measuring every curb. The latest self-thinking car can also handle inclement weather. And it reduces carmakers’ liability.
Machines are learning to think for themselves. This has never happened before. And the world might be forever changed. The upside is that early investors will profit the most...
American Consequences | 41
Made with FlippingBook flipbook maker