I know enough about the software in BMW (NBT/OS7) and Audi (MIB2/MIB3) instrument cluster stacks to know there's at least as much complexity — if not substantially more — in many of the legacy brands. Not to mention the exponential complexity which comes from their highly modularised approach, where systems from a variety of external suppliers have to co-ordinate with each other.
By contrast, the Tesla software stack is (or appears to be based on a few minutes of research) shockingly straightforward considering its apparent complexity. Rather than being a hodge-podge of vendor software, it appears to be Qt-based software running within a Linux environment on Nvidia and/or Intel chipsets. Reviewers routinely praise the screen for being responsive and "iPad like". If there's a bloat issue, it'd be interesting to hear some specifics.
As for your quip about "decades of careful evolution in HMI and TCB design" you might have been right 20–30 years ago.
Isn't the role of the VW MIB2 (as in "Modular Infotainment Baukasten") explicitly the infotainment part of the vehicle (and NOT the instrument cluster, cruise-control, etc.)?
I never had an issue with those, as their reach is isolated (or "limited" as people would say today) to the infotainment part of the car. It couldn't even take control of the climate system back when I had one.
Can't argue much about MIB3, it is just a few years old and a child of the Tesla Software-defined-car era (albeit still tries to uphold Volkswagen's DNA of strictly separating roles of all components, partly making it the mess it is...)
Look at the automotive standards for system verification, in standards like ISO 26262 functional safety. These standards are followed by engineering led organisations that respect safety. Tesla prefers a more laisse faire approach.
This is all orthogonal to software defined vehicles, except that you have to choose to segregate functions to achieve strong non-interference goals, and all that checking might slow down software development and Tesla doesn't like that.
The HMI on Teslas is trash, and drivers are measurably slower and more distracted in simulated and real conditions in their designs. The scan is slower, the affordances are weaker, the multi modal nature causes processing delays. The worst part is design and marketing teams being forced to copy them. Chinese imitations, like the insanely cheap/ugly MG4 tablet instrument, are going to age very badly.
Volkswagen tried to evolve to a more "touch" based HMI -that everybody hated- and is now touting it's abandoning that HMI as the largest redeeming factor of it's cars.
China is banning the ridiculous "innovations" on car handles and further "innovations" on steering wheels.
The Tesla software stack has few advantages: it's cheap and can be easily revised when the Beta-user discovers issues with it. So I have to pause and think to who's benefit it's made the way it is.
From an HMI perspective a Tesla is a nightmare, getting in and out one is constant question as to -why- these design choices were made. Especially after taking out "just doing things different" as a reason. A friend's first additions was loops to the physical door-releases so that passengers could actually get out should something happen and incapacitate the infuriating button-based door releases.
Luckily there is progress such as the recent Ferrari HMI that actually thinks about how the HMI will be used. The central screen even offers a palm-rest for when manipulating the screen. Integrating physical buttons and switches with the canvas of a screen is the logical way forward.
The car industry is soul-searching as we speak on what to do with technology and our interaction with it. But one thing is absolutely certain: whatever Tesla did is not the future.
At least your friend can add loops to the physical door-releases. If the problem is software defined, good luck hacking in even the simplest of one-liner bugfixes.
I agree Tesla door releases are silly, on both sides of the door. As a bald person in a city with hot summers, I am no fan of the glass roof either. But at least I can mitigate those. And neither are anywhere as maddening as being unable to wind windows up after opening the door. Or having to switch the radio off every single f****g time I start driving. While those might sound like mere annoyances, the repetitious inanity is utterly grating. I value physical ergonomics greatly, but there's something about pseudo-malicious software behaviours which make me angrier than any door handle ever could.
On a recent holiday I rented a Model 3 (pre-facelift) for a few weeks. It has a few quirks, but nothing that irritated me. It was an utterly pleasant experience. The quirky door handles became second nature within a day, for example. Navigating maps and music on that screen was less of a driver distraction than in many other cars I've driven. Not perfect, but well above average.
I do appreciate physical buttons, but my 1-series BMW from 2013 has taught me that there's something better than physical buttons. It's having systems behave well enough that the buttons might as well not exist. I almost never touch the climate control. Setting the internal temperature to 22 degrees seems to work perfectly all the time; somehow it always seems to do the right thing. The only intervention I regularly make is to press the "MAX A/C" button when driving home from sports or the gym. And I'm pressing that before I start driving anyway, so it's not a driver ergonomics issue.
By contrast, the Tesla software stack is (or appears to be based on a few minutes of research) shockingly straightforward considering its apparent complexity. Rather than being a hodge-podge of vendor software, it appears to be Qt-based software running within a Linux environment on Nvidia and/or Intel chipsets. Reviewers routinely praise the screen for being responsive and "iPad like". If there's a bloat issue, it'd be interesting to hear some specifics.
As for your quip about "decades of careful evolution in HMI and TCB design" you might have been right 20–30 years ago.