Google's self-driving cars have racked up about 2.2 million self-driven kilometres on actual roads in the last six years, but as impressive as that sounds, it's a pittance compared to what the simulators have been doing behind the scenes. As part of its monthly report on self-driving cars, Google has detailed the simulators it uses to test out software changes for its fleet. Every kilometre driven by a car with sensors has been tabulated as a three-million-kilometre virtual test course for every version of Google's software. Whenever code is changed, it "re-drives" all the kilometres that Google's cars have driven in the real world before, to see how it reacts:
For example, to make left turns at an intersection more comfortable for our passengers, we modified our software to adjust the angle at which our cars would travel. To test this change, we then rerun our entire driving history of 2+ million miles with the new turning pattern to ensure that it doesn't just make our car better at left turns, but that the changes creates a better driving experience overall.
The report claims that Google simulates 4.8 million kilometres of driving every day, which obviously requires some processing horsepower. Luckily, Google has a few data centres to spare.
The simulator is also useful for working out what went wrong in the past. Every time a test driver takes control to avoid a potential accident, Google's engineers later simulate what would have happened had the car continued driving, and obviously fix any deficiencies.
Simulators alone won't make self-driving cars any better in rain or snow. But as a tool to speed up development, and check new versions of software, it's potent. Good thing Google has a few spare computers lying around.