Let me try to make a different example.
The Starship prototype that SpaceX launched last night on a test flight... (if you have not seen the footage, it's here
)
As Eric Berger says in his article
An early version of Starship takes its first tentative steps off Earth
SpaceX built a Mars rocket out of rolls of steel.
In tents.
In South Texas.
In weeks.
There are 2 vital things (that fundamentally are required) they needed to be prior to getting there, the Raptor engine and the flight software.
Elon first started with carbon composite tanks and had expensive tooling made for making the large carbon fiber rings for the body of the Starship.
When early pressure tests failed and no way to increase production speed for their carbon fiber approach, he pivoted to stainless steel (30x variants) and literally scrapped the expensive carbon fiber tooling.
Now, they are cranking out prototypes every few weeks. We will start to see additional hops to smooth out the fueling, launch and landing process further and gather more data.
If you look at SN5 it looks like it has a bunch of appendages on the outside skin. The vast majority of those black canisters will be removed/deleted as the Starship will be using its own fuel to pressurize the tanks rather then using helium (aka Autogenous pressurization). And their RCS (reaction control system) thrusters will be replaced to use the Methane as fuel, further removing the need for additional storage containers and removing complexity of ground operations and maintenance.
Rockets are complex beasts, every system or sub-system that you add to your rocket must be able to withstand the forces during launch and landing... and handle the conditions in the vacuum of space. So, removing sub-systems and using the same fuel as the main propulsion allows them to eliminate a lot of complex testing and software to control the additional sub-systems.
I see the Tesla approach to FSD in a similar light.
They have the 2 vital things the custom designed FSD computer and the "Tesla Vision" software to run on that computer. (we are obv. still waiting on the rewrite)
The rest of the sensors need to provide enough detailed information for those 2 vital things to make useful & safe decisions/predictions while operating the vehicle.
Adding additional sensors that provide more information then is needed by the system is only going to have adverse effects on your entire fleet down the line.
If new sensor slow the the progress (just like carbon fiber did for Starship) it is time to scrap those sensors and rethink your approach.
Each sub-system has it's hardware and maintenance associated with it, then it has it's code and the maintenance that goes with that sub-system to get the useful data out of that hardware.
The "first principle" approach that Elon brings up often, in this application is passive cameras with active radar and ultrasonic sensors. Since they've revealed their vision in 2016, they've stuck to it, even when they announce that they are doing a rewrite of their software the underlying vision and principles of their solution remained the same.
As a throwback to "but Elon said" .... here is an interview with Wall Street Journal from September 2014.
Elon is quoted as saying
"They will be a factor of 10 safer than a person [at the wheel] in a six-year time frame"
Obviously we are at the end of that 6 year window and will have to see what the 4D rewrite brings to see how close or how far Elon time was off.
Also, it is clear that the current implementation of Autopilot is not there.
paywalled link
Tesla CEO Musk Sees Fully Autonomous Car Ready in Five or Six Years
non-paywalled link through Verge
Elon Musk says self-driving car technology still has another 5 to 6 years to go