[UK Discussion] This is what a self driving car looks like

[Zilla91]

That one's not true, evidently. Phantom Braking in the i4 Google has pages of results, and I note that one of its suggested searches is 'Mercedes phantom braking' so I'm going to guess that it's present there too.

I'm going by my experiences of comparing the same drives I did with my X3 and subsequently with my Model 3 and Y. Comparing the two, phantom braking happens often while on the X3 it happened once and even then it quickly realised what it was doing and corrected itself before I could hit the accelerator.

If you're going to use Google as your source, maybe you should compare the number of results for Tesla phantom braking compared to other brands.

Edit: Just followed that link, it looks like the person who reported the phantom braking incident has been tinkering with his car using BimmerCode to alter some thresholds. Read into that what you will.

It will be interesting to see if the camera-based attention monitoring gets good enough that it replaces it or whether it just gets used alongside the current system.

It won't, because the camera can't 1) see the steering wheel and 2) not all cars have the IR blasters so unless you've got a car from 22 onwards then it'll be entirely useless at night.

People comparing 4-5 year old, stale code with what other autos are doing right now are using out-of-date information to base their conclusions on.

See what I'm actually doing is comparing my experiences with my X3 assistance system, which was released all the way back in 2017 if not earlier, and comparing that to Autopilot, which may be a bit long in the tooth but is constantly being updated with OTA updates.

As I said, I've not had much time with the newer BMW assistance systems but if I did, I suspect it'll reflect even worse on Tesla.

 

[bkp_duke]

See what I'm actually doing is comparing my experiences with my X3 assistance system, which was released all the way back in 2017 if not earlier, and comparing that to Autopilot, which may be a bit long in the tooth but is constantly being updated with OTA updates.

As I said, I've not had much time with the newer BMW assistance systems but if I did, I suspect it'll reflect even worse on Tesla.

No, this is false info. Autopilot (the original, not FSD) is NOT being updated constantly. It hit maturity back in about late 2018 and hasn't had any significant updates then.


And I never stated that phantom braking had been solved. I've experienced it in both our cars, although it has markedly improved in the Y over the past year on everything by highways (FSD code base, not AP code base).

 

[Zilla91]

hasn't had any significant updates then.

So you are telling me there were no changes to the Autopilot code last year when it went to vision only?

it has markedly improved in the Y

Might it not have something to do with the height of the vehicle and the camera being able to see a bit more ahead?

 

[WannabeOwner]

hasn't had any significant updates

so you are telling me there were no changes to the Autopilot code last year

That wasn't what he said ...

 

[yessuz]

that BMW was plug-in hybrid. not much of a point of increasing anything range or power there, isn't it?

There is also no point of doing anything else, as all other components work as intended.

It took Tesla just 4 years to make AHB working as in other cars that was made 10 years ago.

here' first google result:

BMW software update improves Digital Key and adds app-based Remote Control Parking

BMW is rolling out a new update affecting approximately 4.7 million vehicles. The latest software makes the Digital Key easier to share and activate and brings an app-based Remote Control Parking feature to several models. Other features include a new mobile operator for personal eSIM and an...

www.notebookcheck.net

let's see:
- Adds remote parking added (while Tesla removed..)
- Adds steering assistance while operating cars with trailers...
and so on.. depending on the update

https://www.bmw.com.sg/en/topics/offers-and-services/bmw-digital-services-and-connectivity/remote-software-upgrade-21-11-22-03.html

but I am sure, all these

• Optimized engine control
• Optimized Hill Descent Control
• Optimizing DC charging
• Electric drive
• Tail lights
• Optimized turbocharger control
• Updates of various control units

is just infotainment and they do not do any "engineered parts" updates...

guys, really?

 

[yessuz]

That wasn't what he said ...

so going to TV (while crapping out AP functionality (reducing max speed in and increasing min distance while making car more janky...) is not SIGNIFICANT?!

 

[bkp_duke]

So you are telling me there were no changes to the Autopilot code last year when it went to vision only?


Might it not have something to do with the height of the vehicle and the camera being able to see a bit more ahead?

From everything I've read (and I read all the release notes that get pushed to our cars), the only thing that happened with AP code was that radar inputs were shut off (vision was already in place and the major input). The actual aglo that is used didn't change. That makes sense because Telsa was using neural net AI programming for FSD, and that was their focus. That kind of programming is not compatible with the old C/C++ software stack of AP. So basically AP was considered "mature enough" and left as-is except for the removal of the low-resolution radar (which was a substantial contributor to phantom braking, but not all of it).

When we got the Y it was just as bad as the 3 in terms of phantom braking. We've seen marked improvements on that, but it still happens on the highway (where the AP stack is running, not FSD). Ironic to this discussion, Elon announced overnight that FSDb (with the replacement to the AP stack) will go out this weekend to the cars with FSD. I have no idea how long it will take for that stack to get to the whole fleet (cars that don't have FSD). And then, there are some early cars like mine with HW 2.5 that may never get that stack (HW 2.5 uses older NVIDIA chips on the boards, which are not code-compatible with the NN code being used in HW 3 cars - those have custom Tesla chips).

 

[mnobleuk]

To put it even more simply, one company allows people to go out in their tech without a driver.

That's not quite true - might be dancing on the head of a pin here but Waymo has remote drivers which it asks for assistance whenever it needs to. As a passenger you don't know when the vehicle has asked for assistance or not - therefore its very hard to compare with something like FSD where it's obvious it needs assistance from a human.

 

[WannabeOwner]

so going to TV (while crapping out AP functionality (reducing max speed in and increasing min distance while making car more janky...) is not SIGNIFICANT?!

From your (i.e. driver) perspective then sure. From a Software DEV perspective (which I take as the basis of the thread) then no, I doubt very much that it was.

I'm a software engineer - no knowledge of the actual Tesla code, of course - I've have had plenty of instances where we leveraged existing code to accommodate something new - some bit of hardware that we relied on became obsolete and was no longer available / changed out of all recognition / whatever. But whenever we have done a complete re-write (as I would categorise FSD) we would be "Park all DEV on the old stuff, and just do maintenance" and then 100% of the DEV effort is on the new product. In part because we needed to get the new product to market to be competitive, and the old stuff is obsolete so we don't want to spend any time and effort on it. Customers might want us to ... but they will be happy when we release our Quantum Leap Version 2 and that's the course that we've chosen.

When we moved from Windows APP to a Browser Based APP for example. 99% of our [Windows APP] code was U/S ... lots and lots of work to create a Browser APP replacement for the old style Windows APP. We weren't going to spend one iota on the Windows APP that we didn't have to

 

[WannabeOwner]

don't seem to put much effort in to making the best car that they can 100% finish right now

Beats me why they don't get a Summer Intern to fix wipers and so on.

But there have been major niggles in the past that got fixed (albeit not in the timeframe folk had wished for). e.g. Waypoints in SatNav ... we waited years and years for that ... actually I think the way that has been done is much nicer than what I had anticipated.

Most other companies (in any industry) would not leave their entire customer base on a stagnant product for 4 years while competitors gain or overtake them to focus on the new version

IME Software companies do ...

I would challenge with: AP is "good enough", and Tesla don't intend to do any significant DEV on it. FSD (if they solve it etc etc) will be a revolutionary, rather than evolutionary, improvement.

The steering wiggle is genuinely annoying

I have (just) my right hand at 4 O-Clock. The weight of my hand / arm provides some rotational torque and I never get an alert from AP

 

[Bouba]

A lot of the anti FSD in Europe is born out of frustration....because we don’t get it, literally...no Beta for us.
To me I think that L2 available to 300 million people (in North America) is far superior to a Robotaxi available to a few thousand people (in North America)...but it still is annoying to be on the outside looking in.
To be honest, Robotaxi isn’t really a lot of use...we are all car owners ! But it would certainly come in handy if I could go to the pub get rat-arsed (a nice colloquialism for our American cousins) climb in to the back seat of my Tesla and it drives me home.....now, we know this is coming...unfortunately it won’t happen before Friday night

 

[Mullermn]

I have (just) my right hand at 4 O-Clock. The weight of my hand / arm provides some rotational torque and I never get an alert from AP

This doesn't work for me. The closest I can get to a brain-free nag avoidance technique is to kinda tap/twitch my hand along with the music. The weight of my hand alone definitely isn't enough.

What setting do you have the steering resistance set to? I've tried light and medium to see if this makes a difference but I couldn't really detect one.

 

[WannabeOwner]

we are all car owners

We could all become ride-hailers ...

My car is 2p a mile to drive for Leccy and probably £1.50 on depreciation! I've had the hassle of one car taken-and-driven-away, waiting months for repair several times, and broken down and stranded for hours waiting for recovery on multiple occasions.

Can't see any upside of owning a car ...

 

[WannabeOwner]

kinda tap/twitch my hand along with the music

I'm way too old ... but that sounds like a perfect millennial solution

Weird how things that work for one don't for another (although in fairness your hand/arm may not be as chunky as mine)

I seriously wonder about the user-to-user variations I read on here (phantom braking, AP weaving in lane, all sorts). I'm a software bod, given that these things exist then in my view Software can do a darn good job of levelling the playing field in such situations (if its in the DEV Spec of course)

I would define myself as a [software] design-engineer. I'm strong in both decisplines, which is a bit unusual. Usually you have architects who design things that look amazing but cannot be built, and engineers who build things that are rock solid but look ruddy awful. I'm fortunate to have a foot in both camps ... so maybe my assumption of what software should do only happens in my workshop

What setting do you have the steering resistance set to?

Is that the Comfort / Sports steering setting? If so I think I have it in Comfort (but drive as if it was in Sports )

 

[yessuz]

This doesn't work for me. The closest I can get to a brain-free nag avoidance technique is to kinda tap/twitch my hand along with the music. The weight of my hand alone definitely isn't enough.

What setting do you have the steering resistance set to? I've tried light and medium to see if this makes a difference but I couldn't really detect one.

i actually hold it just above the 3-o'clock bar of the steering wheel...

or you can try 500g weight ;D

 

[GeorgeSymonds]

They've identified areas where they think the statistics don't tell the whole story. That's not the same thing as the data not being credible. Statistics often have limitations. How would you like Tesla to correct for the demographic issues if (hypothetically) that data isn't even available, for example?

They could start to produce their passive safety data that they decided to drop as a benchmark where the demographics are effectively neutralised. We can then compare the uplift removing demoigraphics, age of vehicle etc. When we look at the historical relationship between pasasive (on all roads) and AP (primarily on safer roads) the ratio is much closer than the all cars comparison. It also measures the events ont he same basis. So just why did they stop producing those statistics?

They could then select comparable scenarios for the active and passive data - for instance metrics with similar road type, in daylight, in dry weather to remove such variables and so we can see the uplift or not in results. That would leave us with some variability such as traffic density but by now we'd have removed many of the other variables.

They could also look at forced disengagements and report those alongside, as any foreced disengagement in a self driving context is trouble.

I think its the wrong question. AP/FSD is flawed. So am I (as a driver, ask my wife about the rest ...). But I reckon Me + AP is a lot better. Both of us are good at spotting things the other doesn't. Now ... at some future date ... I expect that FSD will do all the bits that currently I am better at.

"You plus AP" is the definition of a level 2 system. The question posed is about self driving which is "AP v You",, not "You plus AP v you"

If somebody asked if driver assist systems play a role in reducing accidents when if used corectly then of course they do. But too many are asserting that because AP may or may not reduce accident rates when active in conjunction with a human, it therefore implies that AP alone is safer than humans. Thats not a conclusion you can draw.

 

[bkp_duke]

If somebody asked if driver assist systems play a role in reducing accidents when if used corectly then of course they do. But too many are asserting that because AP may or may not reduce accident rates when active in conjunction with a human, it therefore implies that AP alone is safer than humans. Thats not a conclusion you can draw.

Then the Waymo vehicle in the OP doesn't qualify either. Those cars have real humans monitoring and intervening remotely. They are a far cry from autonomous. Plus, they only work in 2 cities on the entire planet . . .

 

[GeorgeSymonds]

Then the Waymo vehicle in the OP doesn't qualify either. Those cars have real humans monitoring and intervening remotely. They are a far cry from autonomous. Plus, they only work in 2 cities on the entire planet . . .

Waymo has a license to operate at level 4 without a driver in 2 cities. That is not the same as the car is incapable of driving in more places.

How many locations does Tesla have a license to operate at Level 4? None.

How many Level 4 miles has Tesla ever driven? None

How many level 4 miles did Waymo drive in 2021? Over 2 million.

So.. who’s putting their money where their mouth is and doing it, even if it has issues, and who is still doing level 2?

 

[MrBadger]

I don’t know why all this binary one is better than other crap. They are different attempts at the same problem. None work as well as people thought they might. They all have their positives and all have their negatives.

For me, if I could be driven up a motorway hands free, safely and know in advance where I would be needed and where I could relax, that would be ideal.

But someone else will have their own criteria of success.

It’s just a shame that some of the solutions have over promised and under delivered. More fool me for believing the hype at the time and my sticking hand in my pocket. Having worked for a small company that ran L4 at the time, I kinda thought that companies with many times the resources would have been much further ahead than they were, but perhaps not.

 

[bkp_duke]

Waymo has a license to operate at level 4 without a driver in 2 cities. That is not the same as the car is incapable of driving in more places.

How many locations does Tesla have a license to operate at Level 4? None.

How many Level 4 miles has Tesla ever driven? None

How many level 4 miles did Waymo drive in 2021? Over 2 million.

So.. who’s putting their money where their mouth is and doing it, even if it has issues, and who is still doing level 2?

Again you get hung up on the definition of L4 and completely IGNORE that Waymo is effectively and practically worthless. As in Google lost BILLIONS on it and it is NOT a highly utilized service in both SF and Phoenix. No one in other cities are clamoring for it and asking for it.

You equate that Tesla had not applied for these L4 licenses with that they can't do it. That's false. Tesla has intentionally chosen the LESS LEGALLY LIABILE option of using a L2 license to work on L5 tech (which is completely legal in the USA, not so much across the pond).

But hey, you have nothing like it in the UK and you probably have never been to SF or Phoenix used Waymo (I have).

But I see how that makes you an expert.