Tesla has had Autopilot in their vehicles since 2014. And, their vehicles have had all the hardware they need for Full Self Driving since 2016.
There was a further updated their vehicles with what they called Hardware 3.0 or HW3 in March of 2019. It was after that time that Tesla started using what is called Shadow Mode.
What is Shadow Mode?
Tesla Shadow Mode is a mode that is run on the second chipset of their Full Self Driving hardware. It uses a copy of the software along with specific updates or tests to run specific scenarios in the background.
The reason this is done is to do real world tests in simulation only in order to gather data and feed it back to the training system for FSD. This is obviously safer to do than test things on the “real” version of the software and potentially harming drivers or other road users.
By making use of their extensive fleet of cars, Tesla is able to run scenarios and gather data from over a million vehicles in order to make faster improvements to the system.
Shadow Mode is just one part of this feedback loop that Tesla uses.
You can see the head of the AI team at Tesla, Andre Karpathy, explaining an example of Shadow mode in this video.
Hardware 3 and Shadow Mode
Due to the fact that Hardware 3 has two exact copies of the same hardware on the circuit board, Tesla is able to run Shadow Mode for testing at all times. It also serves as a kind of backup in case on part of the circuit board were to fail.