The Truth About FSD?

[Sparkeur]

It would appear that in the USA, the NTSB/NHTSA are starting to push back against Tesla's FSD testing. While the attached article is about the potential dangers of testing FSD on public roads, it also seems to show the difference between the reality (lawyers for Tesla) and the marketing (Elon Musk). I think we are all aware of the current limitations but live in hope that things would change sooner rather than later. Reading between the lines, this article seems to cast a different light on the current and near-term future capability of FSD.


'Musk repeatedly hypes Autopilot and FSD to his massive following on Twitter and in media interviews, but in correspondence with regulators, and in the fine print of Tesla financial filings, the company’s legal team refers to these systems in a more subdued and precise tone.'

A federal agency warns Tesla tests unfinished driverless tech on its users

 

[Dilly]

I can’t bin my driving gloves anytime soon then...

 

[CyberGus]

I am shocked, shocked to see a CEO hyping his company's products.

As a new technology, vehicular autonomy has few regulations and little case law. Greater governmental intervention was inevitable, and as it happens, necessary,

 

[NorfolkMustard]

This is that Greenspan twisting of words isn’t it?

this “FSD Beta” currently on a few cars isn’t being called level 5 or anything close by anyone involved in it. Big warnings on the screen and in their welcome emails about the driver needing to still be in total control at all times.

only some outsiders are conflating the FSD end-state with what’s currently being tested.

 

[AnthonyLR]

I think it is very good of US Customers to be our test pilots for us. By the time it starts rolling out in the UK it will have done very many US miles. Of course driving on the other side of the road, roundabouts and other UK-centric quirks will need quite a lot of work when it does arrive here. FSD Beta does appear to be making rapid and significant improvements

 

[Durzel]

To be honest I think I would be pretty happy with the current carnation of FSD Beta that Americans are getting at the moment.

I can't see cars being anywhere close to being completely driverless for at least 5 years technology wise, and probably another 5 years before legislation catches up.

As for the Robotaxis concept - it's not even worth giving it any serious thought.

 

[Tony Hoyle]

It's still got some very rough edges that can cause accidents.. I'm really not sure it's a good idea to give that kind of beta software to *everyone*. If you have a shortlist of users if they take risks they can be thrown off the programme.

eg.

- as someone posted 'it was going so well... until it wasn't..'

 

[GeorgeSymonds]

I have to say I have a lot of sympathy for the regulators. Musk is clearly seeing every establishment as inept and likes to play games whether the stock market, the regulators, the banks, everyone.. he seems to feel he's invincible and can do what he pleases, even his job title change this week seems to be taking the preverbial.

He's also enjoying the cult status of getting people to blame regulators for holding things back as if they would be much further forward if it wasn't for them. The UK owners group often suggest lobbying MPs about the limitations we've placed on autopilot, but when you look at them they're barely anything - summon across a carpark, tight bends at speed and needing to tap the indicator to acknowledge a lane change. If only they were the only things wrong with FSD at the moment.

There's a known issue that the greater the capability of the system, the more reliant and casual people may come when using it. I think it was Clarkson who said you could reduce accidents by removing seatbelts and sticking a big spike on the steering wheel (or something along those lines). The argument that you may have driven a thousand miles and not needed to do anything is great, but unless youre ready for that moment when you're needed you're at risk of a serious brown trouser moment 10 or so times a year. So how do you keep people attention on the system. A disclaimer isn't going to do that, the world has become immune to such things after somebody sued McDonalds because the coffee they bought contained a hot liquid which resulted in everything being covered with warning labels.

So you mix it all up and he's rolling out software to people that's not finished, may not even be a great drivers aid, those drivers will be the ones blamed if it goes wrong and they lose concentration through boredom or more likely feel checking their emails is worth the risk because the system is "reliable" and the nags are gone of you just hang a heavy arm on one side of the wheel, or a weight as some have done

I don't even get the "next version is camera only". It said to suggest its better and its all thats needed. Do we believe that a pair of fixed cameras in the centre of the windscreen is all that's needed to road the road ahead? I often need to lean my head to one side to get a better look past the car in front or reposition the car on the road to get a better view, whereas driving on the continent can be a nightmare for vision at times.

Give it time and the potential is there for it to offer massive benefits (so I get the vision) but over the last year, the "miles driven between accidents" ratio of active AP to passive/safety systems only AP has already been shrinking (was 1:2.4 down to 1:1.7) with the miles between accidents in Q4 2020 only marginally better than the stats from 2018 when they first published them.

I don't know what the answer is, others use cameras on drivers to watch eye movement, a regular confirmation when a message is displayed at random, both of which will be taken as an afront to many owners. There's a guy on another forum who just doesn't get it, he said the thing he hates about not using autopilot is that he need to stare at the road the whole time and prefers to look away and relax his eyes at times.. well right now, you still have to look with these systems engaged.

 

[Tony Hoyle]

Tesla needs FSD (or 'city driving' if they want to call it that) to be level 2 to avoid regulators forcing it to be licensed. But that means that AP - simple lane following - and FSD - a fully autonomous system making driving decisions - are the same 'level'..

IMO. the 'levels' are meaningless marketing (and legal) fluff. Get the bugs out and Elon will be declaring he's reached level 5..

 

[GeorgeSymonds]

I think thats part of the confusion, a car that does all the driving can be working at level 2 as its still being supervised even if no actual intervention is required, the benefits and real worth comes from the driver being able to disengage from the driving supervision aspect. Musk is hoping to pull off this trick, he's even telling the regulator that no autonomous miles are being driven to avoid the rigour that goes with it and reporting intervention levels. and hopes that he'll have a fab system at some point and then he'll be able to tell the regulator it no longer needs supervision. Its a dangerous game because it will lead to complacency where drivers decide they no longer need to pay attention before they should.

 

[Yev000]

Its a dangerous game because it will lead to complacency where drivers decide they no longer need to pay attention before they should.

George, I can understand your fear. I think you are giving Musk a bit too much of the credit here, Tesla is a big company now and will carry on down the FSD road without Musk if it comes down to it. Too much money to be had being first. When is something ever truly "finished" anyway? And yeah it can potentially kill people with complacency, but no more so than the army of drunk or tired drivers on the road right this second.

Thing is, it's inevitable. There are not a lot of things in history that progressed human kind without being dangerous at first. There are no unwilling participants here - to my knowledge no-one has mowed down people on the street or killed another driver with FSD. A few deaths with AP - mostly due to complacency.

All things being equal I think the regulators are doing more than enough to slow down progress and take the "danger" out (although you could argue they make it more dangerous by restricting AP). We all want to live long and productive tax paying lives after all....

Are you that concerned for the poor conned drivers that use FSD? I'm not sure they are that many of those....

 

[Mm2021]

Really interesting conversation. I have to say scepticism helps us avoid internal disappointment but it is hugely counter productive to technological advancement. We must not be complicit in this FSD related scepticism. Hope and belief is what got us to the moon. There have been very few technological advancements that have changed the world that didn’t have a bumpy start. I know, I know- road safety, lives- of course we would never want to endanger the lives of others or ourselves BUT that is what regulators are for- leave them to it. If we join in with the negativity we will be assisting in delaying technology that will transform the lives of everyone we love generations into the future. Our children’s children should see a future where road traffic related deaths are unheard of. We must hope and aspire to that. Let’s get behind every autonomous company venture that is trying to push the envelope and as a community have our voices heard against regulations that are obviously over zealous and actually hindering advancements that could save lives today not just into the future.

I guess all I’m saying ladies and gents is please let go of the past, of holding on to that Top Gear-esque notion that old is best, driver behind the wheel is superior. Of course it may be right now, but if we are too afraid to take technological risks towards a future that is “order of magnitude” (Elon’s favourite phrase) safer and we wish to get there so slowly that that our children are exposed to the same risks in vehicles we have been all our lives from reckless, thoughtless drivers, drunk drivers, inattentive drivers, unexpected road obstacles etc etc then we should feel nothing but shame on ourselves. We have the ability to help companies like Tesla change the world- so put a smile on your face, look to the future with excitement not disdain or bitter scepticism and let‘s help push the world to a brighter future as soon as possible.

 

[GeorgeSymonds]

I accept fully that the goal is worth striving for. I think I said that, but we can still debate the way we get there. I think its patronising to think the objections to what they're trying to do is because some are tied to humans are better and FSD is not worth obtaining. I don't think thats the complaint at all.

There's a couple of key points:

- The data Tesla shares is meant to indicate positive progress, but its actually not doing so when you look at the figures.
- Incremental steps forward and the willingness to reverse decisions where they transpire to be wrong has to be more important for society that blindly pushing forward to protect current share price until the company can withstand the shock if it fails if thats what may be happening.
- Society as a norm works on the basis that it tolerates people killing themselves through there own actions with laws and taxes meant to encourage better behaviours. We have speed limits with a slap on the wrist but we don't ban the car, we have tax on cigarettes, we don't outright ban them etc.
- Nowhere in peacetime society do we sanction corporate responsibility that can result in death on a large scale with the exception of medicine (and thats balancing two unpleasant outcomes) and sport (things like boxing and motorsport, and not without repercussions). So sanctioning a self driving capability that may be 2, 3, 4 times better than the average driver still results in massive global death toll. Flying is incredibly safe per mile travelled yet a plane accident gets extensively investigated and Boeing have had a fleet of their planes grounded for some time because of a couple of accidents. Per mile they're still way safer than car travel with human drivers. The point is simply its going to create more than a statistical comparison of FSD v drivers to determine if the risk is worth taking, its a change of underlying culture in the risk taking that governments are prepared to take. That may be over due but that goes way beyond self driving, this goes to the values of society and the expendability of some lives for the greater good. If we're happy for 300 a year to die at the hands of a computer program, not because of any action or inaction that person took, just wrong place at the wrong time, then where do you stop?

But lets say they can reach the 9 9's safety threshold, then sure, I'd be happier with self driving cars around me than some we come across, but that doesn't excuse the journey to get to that level as if we were in some state of war or a pandemic. Even hospitals are getting rapped at the moment for DNRs of dying patients on overflowing covid wards during a pandemic and trying to hold them to account.

I've seen nothing scientific about the Tesla approach, the argument for a vision based system seems to be "we drive with 2 eyes" which is not what I'd call a well argued point. So it is very reasonable to ask if the Tesla approach is appropriate and commensurate with the aspirations and time pressures or is it simply they are throwing people into risky situations in an attempt to be the first to fulfil the goal?

 

[Yev000]

I accept fully that the goal is worth striving for. I think I said that, but we can still debate the way we get there. I think its patronising to think the objections to what they're trying to do is because some are tied to humans are better and FSD is not worth obtaining. I don't think thats the complaint at all.

There's a couple of key points:

- The data Tesla shares is meant to indicate positive progress, but its actually not doing so when you look at the figures.
- Incremental steps forward and the willingness to reverse decisions where they transpire to be wrong has to be more important for society that blindly pushing forward to protect current share price until the company can withstand the shock if it fails if thats what may be happening.
- Society as a norm works on the basis that it tolerates people killing themselves through there own actions with laws and taxes meant to encourage better behaviours. We have speed limits with a slap on the wrist but we don't ban the car, we have tax on cigarettes, we don't outright ban them etc.
- Nowhere in peacetime society do we sanction corporate responsibility that can result in death on a large scale with the exception of medicine (and thats balancing two unpleasant outcomes) and sport (things like boxing and motorsport, and not without repercussions). So sanctioning a self driving capability that may be 2, 3, 4 times better than the average driver still results in massive global death toll. Flying is incredibly safe per mile travelled yet a plane accident gets extensively investigated and Boeing have had a fleet of their planes grounded for some time because of a couple of accidents. Per mile they're still way safer than car travel with human drivers. The point is simply its going to create more than a statistical comparison of FSD v drivers to determine if the risk is worth taking, its a change of underlying culture in the risk taking that governments are prepared to take. That may be over due but that goes way beyond self driving, this goes to the values of society and the expendability of some lives for the greater good. If we're happy for 300 a year to die at the hands of a computer program, not because of any action or inaction that person took, just wrong place at the wrong time, then where do you stop?

But lets say they can reach the 9 9's safety threshold, then sure, I'd be happier with self driving cars around me than some we come across, but that doesn't excuse the journey to get to that level as if we were in some state of war or a pandemic. Even hospitals are getting rapped at the moment for DNRs of dying patients on overflowing covid wards during a pandemic and trying to hold them to account.

I've seen nothing scientific about the Tesla approach, the argument for a vision based system seems to be "we drive with 2 eyes" which is not what I'd call a well argued point. So it is very reasonable to ask if the Tesla approach is appropriate and commensurate with the aspirations and time pressures or is it simply they are throwing people into risky situations in an attempt to be the first to fulfil the goal

So if I'm reading you correctly, you want them to stop and/or shutup about FSD because we, as humanity, have more important things to do right now?

If you were in charge of Tesla FSD program what woud you do?

 

[GeorgeSymonds]

So if I'm reading you correctly, you want them to stop and/or shutup about FSD because we, as humanity, have more important things to do right now?

If you were in charge of Tesla FSD program what woud you do?

That is 100% not what I am saying

I am saying they need to define a sensible and safe approach to development and testing not chuck stuff into the public domain and use enthusiastic armatures to find the flaws and only ban a few who seem to take liberties after the event. be more honest about the risks and their real performance if they are expecting the public to do their debugging to try and head off some of the misconceptions that its better than it is based on a few hand selected YouTube videos (I think we have a right to know the real performance if we're the ones testing it). be prepared to implement things to increase the safety during the transition rather than hide behind a disclaimer (cameras in cars are used by other makes to monitor the drivers attention for instance, my wifes car starts to nag if you look away from straight ahead after 3 seconds, not wait 3 months and hand pick a few who are abusing it, yet we've no terms of engagement and what constitutes proper use etc), maybe even mandate a online training programme for owners before allowed access to the latest code accepting that not all drivers can be proven to have had the training, it would still be a start, and engage constructively with the regulators to understand what's needed to actually reach the goal.

Doing the above would take the gloss of it maybe, and Musk wouldn't be seen as quite the cool maverick he likes to be, but I think ultimately he's either going to risk reputational issues if there are more accidents like the police car one yesterday through carelessness by the driver using the software which will end up undermining the benefits. and/or never get it legally approved which leaves it in limbo land.

 

[MrBadger]

May be of interest the UK Roadmap

Official link UK Connected and Automated Mobility Roadmap to 2030 by Zenzic
Quick unofficial link https://fncdn.blob.core.windows.net/web/1/root/zenzic-roadmap-report-2019.pdf

I previously worked with some of the contributing people on the above so know them well. All highly regarded. There is another roadmap that I have seen from some of the same people, but I am not sure that it is in the public domain yet so will not post it. But it adds more perspective to the above and what is achievable in closer time frames.

 

[browellm]

Hope and belief is what got us to the moon.

Actually it was a lot of really talented scientists, engineers and pilots.

 

[pdk42]

Shock, horror - Tesla FSD is nothing of the sort! Well who'd have worked that one out? Seriously, for anyone who owns a Tesla in the UK, it's patently clear that it's a long, long way from being able to reliably drive the car without the driver constantly monitoring it. I've seen videos of the new beta, which is certainly a step up, but it's clear that the programming is heavily optimised for US road design; and of course it's all relying on rather limited sensor input (in particular, no Lidar). I wouldn't pay £680 for it, never mind £6,800.

 

[Peteski]

May be of interest the UK Roadmap

Official link UK Connected and Automated Mobility Roadmap to 2030 by Zenzic
Quick unofficial link https://fncdn.blob.core.windows.net/web/1/root/zenzic-roadmap-report-2019.pdf

I previously worked with some of the contributing people on the above so know them well. All highly regarded. There is another roadmap that I have seen from some of the same people, but I am not sure that it is in the public domain yet so will not post it. But it adds more perspective to the above and what is achievable in closer time frames.

Looks like an interesting read. Being an automotive engineer I always suspected FSD would take much longer than I think Elon was hoping a few years ago. Neither the hardware or software are really up to the task and I haven’t personally seen much progress over the last 4 years and non whatsoever in the last 12 months.

The closer you get to true hands off FSD the harder the task becomes and the more robust the system needs to be. Humans are incredibly adaptable and will deal with all manner of unpredictable edge conditions. We are also very good at predicting how various other road users are likely to behave and adapt accordingly. Machine learning may be able to replicate much of this experience, but only when the tech is able to identify specific threats, like a bunch of teenagers in a stolen hot hatch!

It would all get much less complicated if road infrastructure was consistent and humans were not allowed to drive alongside autonomous cars. Better still if all autonomous cars were connected. That probably is the true end goal in the next 50 years. Fully autonomous driving zones where you literally log your car into the zone network. This would probably be easiest to implement on motorway networks and maybe inner city loops. But first we have to get rid of all the millions of legacy vehicles.

 

[Peteski]

Shock, horror - Tesla FSD is nothing of the sort! Well who'd have worked that one out? Seriously, for anyone who owns a Tesla in the UK, it's patently clear that it's a long, long way from being able to reliably drive the car without the driver constantly monitoring it. I've seen videos of the new beta, which is certainly a step up, but it's clear that the programming is heavily optimised for US road design; and of course it's all relying on rather limited sensor input (in particular, no Lidar). I wouldn't pay £680 for it, never mind £6,800.

It still surprises me how many people buy FSD without giving it much thought.