2019.20.1 - Just got this in the UK...

[WannabeOwner]

I can't help thinking it needs someone at the highest technical level to really get a grip on their poor software QC.

its just a methodology - well understood in the computer software industry, but not bothered with by the majority - worst culprits being projects (now) led by marketing rather than IT.

No software is perfect, however cost-to-fix goes up in leaps and bounds according to how early in the process you find a flaw - Fixing something at the Design Phase costs almost nothing ... fixing it after release is astronomic by comparison, and also very high probability that the fix breaks something else, because by then everything is interdependent.

here we worry about how we lay out our code, and our naming convention. This isn't just so that it is easy for the next person to modify it (although making that easy reduces the bug-chance too), the Coding Style we use is constantly reviewed in terms of using a presentation that is more likely to get DEVs to spot errors early - i.e. making the typical typing mistakes that people make "smell wrong"

Tesla could adopt the methodologies known to reduce bugs. The Aero industry has a very low bug count (notable exceptions, but "in the main") and so, in fact, does the software Tesla uses to control the car ... so why, then, is their Infotainment and BackOffice such shoddy quality by comparison? Just get the Auto-Software crew to trot round to Infotainment and get them to sharpen up their act.

I'm talking to myself though, Elon isn't listening. he needs to sleep on the back office floor, instead of the assembly line floor (or camp in this forum ...)

 

[Peteski]

Tesla could adopt the methodologies known to reduce bugs. The Aero industry has a very low bug count (notable exceptions, but "in the main") and so, in fact, does the software Tesla uses to control the car ... so why, then, is their Infotainment and BackOffice such shoddy quality by comparison? Just get the Auto-Software crew to trot round to Infotainment and get them to sharpen up their act.

This is so true. At least the basic car control software has proven to be pretty robust. At least I'm only having to reboot the infotainment system on a daily basis while the car itself keeps on going fine. However I have had a couple of control system issues, but nothing serious. I had a scare once when I got in the car and it refused to select drive (either forward or reverse - just remained in P whatever I did). That was cured by getting out, locking the car and then starting over again. Never to be repeated since, although I've been presented with a dead screen on a couple of other occasions requiring the same lock, unlock procedure to wake up.

 

[Mark_T]

The policy of 'continuous improvement' in the hardware creates the worst possible target for QA testing as well.

Much like Android hardware it is literally impossible to test every combination of sub-systems out there for cross-compatibility issues.

The fact that we don't all get the same bugs points to this as well...

 

[WannabeOwner]

The policy of 'continuous improvement' in the hardware creates the worst possible target for QA testing as well.

Yes, it makes the job far worse, I agree. I would want a simulator for all that, but I don't know how feasible that is. But:

• Changing the graphics equaliser from 3-sliders to 5-sliders and it then forgets the setting every time you PARK
• Rearrange the screen and "forget" to put the CHARGE NOW button back on scheduled charge
• Update causes USB Music playlist to change from sorted-order to random
• and so on ...
• Also errors fixed in one version, back in the next (failure on regression-testing)

These are QA-101 software testing failures in my book ... and they worry me in case that sloppiness transgresses from Infotainment team into the safety-critical software components

 

[squarepeg77]

>> The Aero industry has a very low bug count

Yeah, but safety critical standards are onerous and expensive and still you get errors like Boeing’s that cost hundreds of lives.

You’d get probably one Tesla update a year if they went or the same lengths for their infotainment systems as for the control software

 

[gangzoom]

I can't help thinking it needs someone at the highest technical level to really get a grip on their poor software QC.

I was actually thinking the exact opposite yesterday.

Stuck in traffic, AP doing most of the 'driving', Spotify streaming music, Browser actually working (MCU 1, browser came back to life with 2019.18), and map showing me how much more traffic there was to go.

The amount of software integration needed must be crazy, compared to what anyone else is doing Tesla is leagues ahead.

I've also not had to reset the MCU for a good few months, even Spotify 'hangs' has pretty much stopped since mid March, even the slight delay on jumping between the settings options on the MCU is now gone. Tesla are clearly optimising the code still even for MCU 1 cars.

Our 2017 build X is now more functional than when we bought it and continues to improve. On the other hand our Lexus saloon is forever stuck with its awful infotainment system that came from the factory forever!!

 

[WannabeOwner]

still you get errors like Boeing’s that cost hundreds of lives.

Inconsequentially small risk ... first figure I got out of Google was 32m airline departures in 2013, 90 commercial airplane accidents, nine with fatalities, total of 173 people

... but I think that particular bug a) maybe could have been caught during DEV/QA b) Boeing certainly should have reacted sooner as data started being reported, before the final fatal crash that grounded the fleet. Bit surprising in Aero industry that that didn't happen (I have Software DEV chums in aerospace industry and always interesting to chat with them about our own bug-mitigation methodologies)

... but might be the result of cosy collusion between Boeing and Apporvl Agency ... those types of issues are a treat.

expensive

Not sure I agree with that.

The cost of fixing a bug, once deployed, is horrendous. Boeing's bug has grounded the whole fleet. For months. I expect some of those Clients will be saying to Boeing "Your fault, here's the bill"

Once automated testing is built it "just runs". Ideal scenario is "the daily-build". Every day a complete system is built, and regression testing run - e.g. all this unattended, automated, usually overnight.

So tomorrow I get a Test Failure report for some change I made yesterday. The work I did is still fresh in my mind, extremely high probability that my fix will a) work and b) have no consequences. If that error is not found until later in formal testing there is a much higher chance that my Fix breaks something else, because of inter-dependencies and also because it isn't fresh in my mind

If the darn thing breaks after release then:

All hands o the pump
Forget about all promised delivery dates for future features

This is all bog standard Computer Science 101 ... trouble is too many projects have now moved from IT to marketing, who know none of that stuff and their priority is on "Pixel Perfect" Presentation and meeting marketing deadlines, not quality ones and IT would say "We can pull that feature and ship, or push launch back"

hence my view that Software Quality is not expensive. (unless you aren't planning to fix the bugs at all ... Computer Virus Authors have it so easy - no after sales support )

 

[WannabeOwner]

I've also not had to reset ...

its now a goodly long time since V9 launch .. and all the features for that which were pulled from Launch are (probably) in the code now ...
... so it is reaching maturity.

It will soon be time for Elon to Launch V10 and take us all back to "nothing works properly"

 

[Peteski]

I was actually thinking the exact opposite yesterday.

Stuck in traffic, AP doing most of the 'driving', Spotify streaming music, Browser actually working (MCU 1, browser came back to life with 2019.18), and map showing me how much more traffic there was to go.

The amount of software integration needed must be crazy, compared to what anyone else is doing Tesla is leagues ahead.

I've also not had to reset the MCU for a good few months, even Spotify 'hangs' has pretty much stopped since mid March, even the slight delay on jumping between the settings options on the MCU is now gone. Tesla are clearly optimising the code still even for MCU 1 cars.

Our 2017 build X is now more functional than when we bought it and continues to improve. On the other hand our Lexus saloon is forever stuck with its awful infotainment system that came from the factory forever!!

I don't have a problem with the UI design, responsiveness or functionality (except for the woeful browser performance). But IMHO as an end user the software QC is still pretty shoddy and there is no excuse for that. As an example, Spotify hang-ups have plagued this software for much of the entire 18 months I've owned the car and it was hardly a new feature back then. Every new software level also generates new bugs almost without fail. The latest one being random switching off of seat heaters. There are not that many vehicle configurations to test and my car is not exactly a rare one (early 2018 MX 7-seater, PUP, MCU 1). Simple things like Dolby Surround randomly switching itself off is just sloppy code. I could easily list a dozen other software bugs in the current release. Individually these bugs don't tend to bother me much (Spotify being a major exception), but collectively they give the impression that the software is written by a bunch of spotty teenagers without a grown up keeping an eye on them!

 

[arg]

Boeing's bug

Not wishing to dispute any of the points you are actually making, but everything I have read suggests that the Boeing (737MAX) problem is not a bug as such - everything was behaving as designed and the behaviour was known.

So the problem was one of specification, and issues around the assertion that this was not a safety critical function (hence not requiring redundancy), and the assumptions around documentation and pilot training to back up that assertion. And also the separate but significant question of whether the feature should have been there in the first place, being essentially a cost saving one at the expense (it has turned out) of safety.

Lots of potential lessons there for Tesla (and indeed for anybody doing engineering, which is why I devote time to reading air accident reports), but the lessons from this particular set of accidents are more around engineering management and standards rather than software QA.

 

[Mark_T]

Today was a good day to test this version with NoA...

Drove from the Oxford SC into Vauxhall, down the M40 for most of the journey. Then through the centre of London and back up the M1.

Sadly it is a fail.

The good parts were the handling of exits, and the general smoothness of the ride with no phantom braking.

Although the exits were handled well, the system indicates far too late for my liking, leaving it right up to the point where the exit lane begins making it an indicate and immediate manoeuvre instead of an indicate just past the final marker then a manoeuvre.

The rest of the lane control was fine in terms of when it wanted to change lanes, just that it is far too slow at actually doing it, so people 'fill the gap' that it needs to over-take well before it actually executes the manoeuvre.

It needs to get a lot quicker at this before it is going to be useful.

 

[WannabeOwner]

The rest of the lane control was fine in terms of when it wanted to change lanes, just that it is far too slow at actually doing it,

On AP2 when you signal to change lanes:

How often does it do it immediately (not sure what is acceptable - 1-2 seconds?)

How often does it not move such that you take over (I'll guess 5 seconds for that, shorter if traffic is approaching)?

On AP1 I've never managed to work out what makes the difference between it changing lanes ... and not. Some times it changes lanes every time, promptly, for a stretch; some times it will go-left but not go-right; and some times it is so slow I take over pretty much every time.

I don't really have any sections where it is, say, 50:50, it is either behaving, and will repeat that behaviour in a couple of minutes time ... or it isn't, and then 20 or 30 miles alter it changes from will-do to won't-do, or vice versa

the Boeing (737MAX) problem is not a bug as such

Yes, I fully agree, sloppy terminology on my part.

It seems to me that there was data enough that Boeing should have realised that it needed improving - as you say that might have been training, but given they have revamped the software as the Fix (if I have that right?) then they could have elected to do that once the data started coming in.

It seems to me that the more sophisticated the automation becomes, the more it is beyond the wit of man to oversee the Cause/Effect elements.

 

[Mark_T]

I think the problem with the lane changing is mostly that it waits too long after a car passes you before it will begin the change.

In that time, someone behind you has now pulled out to get past...

If there is little traffic around it doesn't matter, but when there is a steady flow it stands zero chance of actually making a lane change.

 

[DJP31]

I’ve driven from Surrey to Manchester so M25/M40/M6 and used Nav on AP as much as I could. I found it pretty impressive in the suggested lane changes for overtaking/moving back, traffic was moderate/moderate to heavy.

I have it set to Mad Max and it executed all of them well. It also positioned me nicely to leave the M40 for the Oxford Services but took the exit before. I’m not sure if it was my mistake as I indicated and the car moved across, are you supposed to indicate or does it do that for you just prior to taking the exit?

Only had one bridge phantom braking, and a couple of time it bailed on NoAP when the rain was heavy.

Manchester to Glasgow tomorrow, will see how it gets on with that.

 

[Alset2go]

Got this update the other day but still waiting on Map Update for NOA.

Full strength Wi-fi.

 

[vitesse]

Do loaners from Tesla get OTA updates? Still stuck on this...

 

[Mark_T]

I’m not sure if it was my mistake as I indicated and the car moved across, are you supposed to indicate or does it do that for you just prior to taking the exit?

That one is on you

When leaving, the car will indicate and move without your input.

If you indicate then it will execute a lane change just as it usually would.

I may have to try 'Mad Max' and see if that makes enough of a difference.

 

[DJP31]

That one is on you

When leaving, the car will indicate and move without your input.

If you indicate then it will execute a lane change just as it usually would.

I may have to try 'Mad Max' and see if that makes enough of a difference.

Yep, NoA 1 - 0 @DJP31

Tried it again and it worked perfectly, well not quite, it wobbled slightly on exit before it picked up the inside exit lane.

 

[Mark_T]

Tried it again and it worked perfectly, well not quite, it wobbled slightly on exit before it picked up the inside exit lane.

It does get a bit confused on exits that widen into two lanes as well, that needs a bit more work.

Currently tracks the left line then swerves to the right as the lane divides.

 

[gangzoom]

I’ve driven from Surrey to Manchester so M25/M40/M6 and used Nav on AP as much as I could. I found it pretty impressive in the suggested lane changes for overtaking/moving back, traffic was moderate/moderate to heavy.

Found the same, Leicester to Kendle, 19.20.1, it could be just once you go north of Lunton people relax a bit with their driving .