3 ms·
Google has done an end-run around the challenges that make something like the DARPA Challenge difficult. Driving an autonomous vehicle in an unknown environment
by ef4 12y ago
Google has done an end-run around the challenges that make something like the DARPA Challenge difficult. Driving an autonomous vehicle in an unknown environment is still hard. But that's not the problem Google is tackling.
They're solving autonomous vehicles for known environments that are mapped down to extremely high resolution. Once you have that solid ground truth, adapting to changing conditions on top of it is far easier.
Which is why they've already logged 1.1 million kilometers of autonomous driving, including city streets with high complexity.
What the more traditional autonomous vehicle researchers didn't/don't appreciate is that it turns out to be easier to make the whole world[1] into a well-known environment than make a car than can handle unknown environments.
[1] more precisely, the parts of it that the average person would actually drive on, which is much smaller and simpler.
- ghaff 12y ago>it turns out to be easier to make the whole world[1] into a well-known environment than make a car than can handle unknown environments. I fully agree. And it's the reason we could probably see fully autonomous operations on designated sections of limited access roads in the relatively near future. The unknowns in that environment--mostly other vehicles--are relatively well-defined. The reason for the skepticism about more general self-driving without even a human backup available is that even if the roads are precisely mapped the environment in a city is still largely "unknown." (Although obviously some cities are more complex than others.)