The Google Blog just issued an update about their self driving car program. Their computer controlled cars have logged over 300,000 miles without an accident. Now members of the team are going to use the cars to do some limited commuting, with only a single human within the vehicle. Previously the humans were in two person teams. This is very remarkable for a number of reasons.
1. Image recognition is very hard, and so I guess video recognition is also hard. That said, there is probably a reasonable subset of features that are critical for the computer to recognize, for example. A person can easily tell the difference between a horse, or a cow, or a deer is in the road, which the computer will find difficult. However to avoid a collision, we do not care what the large obstacle is, we just need to know that it is in our way, or on a path to be in our way. The computer is probably much faster than a human in responding to an identified obstacle.
2. This is a very bold move by Google. They started out as a search engine company, and now they do so many things. I try to avoid hyperbole, but they are probably the most innovative company around. Now it might be that they are just the best at buying other companies of interest, but this foray into big hardware is significant. For Google this is a high level of diversification.
The idea of boldly going after an innovation is inspiring. Now I do not know about the self driving car niche, so I do not know how much of the techniques and algorithms are borrowed from academia etc., but Google is most certainly picking up patents by the basketful as they are figuring out how to make this work. There is a lot to be learned by putting yourself out there and making the effort. At some point you have to stop talking and thinking and start making something. I would love to see the full story about the prototypes and simulations and failed tests that went into this effort.
Also, given human nature, I wonder when people or the regulators will feel safe about people giving control over to the driving computer in even larger numbers. At what point do we determine that it is safer than a human driver? I am certain that we could achieve that pretty easily, but it still might have its flaws, and would we require it to be perfect? It is certainly an interesting question.