The Autonomy Loop
It is 1986, you are behind the wheel of a Mercedes truck that cost 749 million euros to build. The truck has driven itself from Munich to Copenhagen and back and you only had to touch that wheel a few times in the whole trip. You are convinced that autonomous cars are a solved problem.
A team of talented engineers come in with an autonomous driving business, the potential market is huge, your friends have been investing and you fear missing out the venture of the century. You decide to invest.
It’s been 30 years, 200 companies, 150 billion dollars invested, 0 autonomous cars. How has this technology been deceiving investors, test drivers and the general public? It is because difficulty is not a number but a distribution.
Building a robot that can drive in a highway, can be accomplished by a group of talented children with five years of experience on Lego Mindstorms. That Mercedes achieved speeds of 170 km/h before you were born.
A car going at 70 mph requires 300 ft to stop. The car receive images 10 times a second, the image then has to be interpreted, and the car has to decide what to do, all of this can take up to 400ms. That’s adds another 40 ft. You need to recognize this specimen 341 ft away to leave space for his luggage. And no, LIDAR won’t help you, even though that thing shoots a million lasers a second, you would be lucky if a single on reaches him.
Don’t get me wrong, what the computer vision community has achieved in the last few years is remarkable. They require millions of examples of highways without pedestrians and millions of examples of highways with pedestrians. You tell me how we can collect millions of the latter.
So that’s the loop of life, with our mindstorms expertise we achieved 99.99% autonomy, more than enough to seduce the public and wealthy people. The reminder 0.01% is pretty damm hard and despite the progress no body thinks 0% is possible. Some ballsy executives thought it was good enough, one of them is in Jail, a pedestrian killed in Arizona, and autopilots are aiming at fire trucks.