Contrary to many other car makers, Tesla is not limited to data from test vehicles as it has access to data from the tens of thousands of Tesla cars that are driven in real-life conditions by real people. Every ten hours Tesla receives about one million miles worth of data that can be fed to machine learnings algorithms to improve the technology or to test new algorithms.
In total, Tesla has collected over 780 million miles of driving data. Anderson also revealed that Tesla tests new algorithms in real-life driving situations via software that runs as an inert background process (and thus never actually controls the car):
Tesla also uses its cars' data connections to try out new algorithms on existing, real-life drivers. The company quietly deploys software to its cars, which then gets tested out in the real world. While that may sound incredibly shady and dangerous, Anderson repeatedly mentions that the software is “inert” and never actually controls the car – it’s just running in the background and sending its results back to Tesla.
And this approach, which is also touted by the company’s PR reps, has propelled Tesla to the forefront of the autonomous vehicle field. While Google, GM, Apple and others are limited to only using data from their existing fleet of test vehicles, Tesla is continuously learning from cars that are being used all the time, by real people.
Source: Neowin