An average person in the US drives some 15,000 miles per year. To drive 3 million miles would take that person 200 years!
So, who's driving 3,000,000 miles per day? The Google car. Not the real one but one that is simulated in Googles data centres.
I was reading the Google monthly report (January 2016) and I was really impressed by this figure.
Google has 55 self driving cars, 22 being modified Lexus and 33 prototype cars like the one in the photo. According to the report they drive some 10-15 thousands miles per week (that is about 300 miles per car per week) in an autonomous way (without any human intervention).
What I found most interesting is that Google store the records of each drive, including the images picked up by the car cameras, the radar information and the data from the many sensors equipping the car and use these to simulate the cars behaviour as their software gets updated.
In practice any new software change (and it happens continuously) is first tested by simulating the autonomous driving on real road situations, the one that were encountered by Google cars in the 1,419,672 miles they have autonomously driven since 2009.
This result in simulated cars "driving" inside Google Data Centres for the equivalent of 3 million miles a day! The amount of processing capacity is staggering, it means driving 35 miles every second. If you want to see this in another way it means autonomously driving a car in a urban/rural environment at a speed of 204,255 km per hour! This should really give you an idea of the massive computation capability required... (these figures are mine, provided to illustrate the point, you can do your own calculation based on the figures provided in the report).
Of course, the simulation is flexible and engineers can also change a few parameters to create a sort of virtual environment to test specific reactions of the car.
A self driving car is amazing, looking at what is behind the scene is even more!