Will FSD be a reality within the Model 3's lifetime?

[pdk42]

I've been pondering FSD. The question I keep returning to is whether FSD will happen, at least within Europe, within the lifetime of current cars. It seems to me that even if the tech can deliver, the implications from a legal, regulatory, insurance and road layout perspective are formidable. If the car assumes all responsibility then who is responsible when the inevitable accident happens? Is the insurance covering the driver, or the manufacturer? How do you mix autonomous and human-driven cars within the same environment? It would probably be better to have separation of the two types, but how could that be arranged on our congested roads?

I know - questions, questions... but it's the weekend!

 

[Jason71]

You should make it a survey. FYI I vote no.

 

[M3HypeTrain]

What do you define as a M3 lifetime? 3 years , 5 years 10 years?
Also keep in mind the FSD Package is not just FSD, there are other features that will be released

 

[Tam]

...whether FSD will happen...

According to Tesla's definition, it will happen as supervised FSD with the upcoming:

1) Traffic lights and signs automatic compliance
2) Automatic driving in the city

which will be activated by the end of this year.

However, for the literal definition, with no drivers, I doubt FSD will happen any time soon. It may be a decade or a few decades.

...How do you mix autonomous and human-driven cars within the same environment? It would probably be better to have separation of the two types, but how could that be arranged on our congested roads?...

Robots needs its own environment to work flawlessly. Current road infrastructure is not optimized for robots. It's best that there's a lane or two designed just for robots (FSD). It will be expensive. If there's no space, you can go above as an overpass or below as Boring Company.

...If the car assumes all responsibility then who is responsible when the inevitable accident happens? Is the insurance covering the driver, or the manufacturer?...

The question was raised during this year during Autonomous Day and Elon said "Tesla, probably Tesla".

By default, whoever owns the property is the one who is liable. Just like if someone dies because of a malfunctioned automatic elevator. The owner of the building is liable but the insurance can pay for the monetary damage. However, the owner might be held criminally due to skipping the cost of maintenance, repairs, and inspection...

 

[Tony Hoyle]

FSD will cover more things over time. Reglulation will lag but not by as much as you'd think (UNECE 79 gets an update at the end of September for example, and the proposed changes cover most of what Tesla is doing now).

Traffic light and sign recognition are 'soon' - so less than a year hopefully? That'll make a huge difference (the next big one being roundabouts but americans don't really know what they are so Tesla won't be working on them).

'robotaxi' is years out. But we'll get something pretty good long before that.

 

[Fullerene]

Have to wear a UK hat for this one - we have form with the Red Flag Act Locomotive Acts - Wikipedia

My dream is I will go to sleep in my car and wake up at my destination, without relying on a human to drive me. This may happen eventually but I can't see it happening very soon.

Given the fact that legislation is always behind the curve on technology and the limiting factor at the moment appears to be legislation then it may take a while.

Considering the recent active UK clamp down on electric scooters, illegal to use on anything other than private land, along with e-skateboards, not directly legislated against but the police have been handing out cautions to people in London recently. Properly insured licensed etc to avoid complete chaos, these make sense as a form of transport. But rather than show a progressive attitude or implement sensible trials, our government appear to have the red flag act mentality. Personally I think the hire & go monetisation attempts of e-scooters causes a lot of problems where they have been allowed a foothold. Private ownership means you'd have at least some respect and not be so inclined to throw it into a canal.

It's all step at a time stuff - the more confidence is achieved the more things will be allowed, the polluting liars (Incumbent manufacturers) will lobby against this sort of thing as they have been caught napping and would rather lobby and put out concept adverts to avoid portraying themselves as anachronistic environmental criminals.

In the meantime the insurers will likely deem non-intelligent cars appear to be less safe as they are wholly reliant on the person in the driving seat. It's a hybrid system, the car should make things safer and the driver is wholly responsible in a legal context, as far as I am aware at least.

I bought FSD in the same context that I got TSLA shares and selected the tow bar option, I'm supporting the cause hoping it will come good one day.

 

[m-i-l]

I think it is fair to say that most people with a reasonable understanding of the technology believe that anything you could call "full self driving" outside of a marketing context is many years away.

That said, Tesla has one massive advantage: machine learning is largely about data, and with the "mass market" Model 3 Tesla will theoretically be able to gather way more training data via their "shadow mode" than their competitors.

 

[Rooster6655]

The software will be perfected given time, shadow mode is underestimated and running simulations on existing data is very underrated.

If say they use the data for the past year and run their simulations on this they can compare their latest FSD software to what real drivers would do and can continue to do this with more data, when the FSD software can choose the right decisions and also avoid any mistakes recorded they will be in theory 'statistically better than human drivers'. I would not be surprised if they have already accomplished something like this but they will face new situations everyday with the new data collected so its never going to be perfect, I think being able to deal with problem situations is the biggest hurdle.

Regarding regulation I would expect selected countries to begin allowing it first within a small area and then expand to larger areas, I don't expect FSD to be allowed in a whole country for a lot longer than the initial trial periods. The time period I would expect is 2-3 years for trials such as a version of a auto taxi after manned trials, for the public version of FSD hands off then 3-5 years but likely to be restricted to where it can be used.

 

[Artiste]

To me FSD means Level 5 autonomy with no geographical limitations. In the UK I can’t see that in the next decade.

 

[Xenoilphobe]

Don't most europeans keep their cars about 8 years. I think you might see level 5 in your next car. By then it will have been renamed 15 times to confuse buyers on what is actually in the car as we have seen since AP1 was introduced.

 

[elecTED]

FSD is already a reality today!

FAKE Self Driving

 

[pgkevet]

To me FSD means Level 5 autonomy with no geographical limitations. In the UK I can’t see that in the next decade.

2 teslas meet on the 3 mile single track road to my house. There is no mobile phone signal but one has to reverse 100yds to a possible overtaking bay and judge the condition of the surface near that farm gate in a snow drift. Not going to happen.

 

[MrBadger]

If Tesla was more into full CAV (connected autonomous vehicles) direction that most of the rest of the industry is heading, this would not happen as I do not believe Tesla currently have the hardware needed to do this although it may not be difficult to retrofit.

 

[Roy W.]

2 teslas meet on the 3 mile single track road to my house. There is no mobile phone signal but one has to reverse 100yds to a possible overtaking bay and judge the condition of the surface near that farm gate in a snow drift. Not going to happen.

I completely agree. The two cars would just sit there looking at each other

They may achieve level 4 in the USA within 5 years, with just very rare fringe cases excepted, but for full no-steering-wheel level 5 my guess is 10 years.

Of course I’d love to be proven wrong

I’m 60 now, so in 10 years I’ll be 70 (see, brain still in tip-top full working order ) and I’d love to have the Tesla drive me around as I become less able. A little afternoon nap in the car (I almost said at the wheel, but see paragraph two...) will be lovely.

 

[Yev000]

Depends on 2 things:

1. Life of the car
2. What you think FSD will do

Lets be clear, I do not see robo taxi happening in UK within the next 10 years.

FSD - as in navigate on AP while you monitor it alertly (no sleeping or texting) - shmaybe in 3 - 5 years in UK.

 

[pgkevet]

..

I’m 60 now, so in 10 years I’ll be 70 (see, brain still in tip-top full working order ) and I’d love to have the Tesla drive me around as I become less able. A little afternoon nap in the car (I almost said at the wheel, but see paragraph two...) will be lovely.

I'm 70 and surely hope this happens. But you're thinking small...I'll be upgrading to the Tesla Panel van with a couple of mattresses for a proper doze and rent it out to perverts when not needed by me (subect to robotic sanitisation between customers), Once my fleet is up and runing it'll be a Robotaxi killer.

 

[Avendit]

Have a read of The Artificial Intelligence Revolution: Part 1 - Wait But Why.

When it comes it will feel very sudden, especially for tesla's with the ability to do OTA. Its very hard to connect the investor day experiences with the reports of FSD still not stopping for lorries crossing and similar, but that is parallel dev for you.

I think Tesla will be able to wring the full capabilities of the current algorithmic approach and hardware in 2-3 years. Whether it is enough to handle the single track road situation above, I'm not so sure. That feels a step beyond the current approach of software created (ie too complex for us to specifically program) if-then-else statements + vision to recognise instructions towards more general smarts and figuring things out for its self from first principals, rather than relying on training.

What could be trained is better awareness of its capabilities so it can reliably drive me to the start of the single track road then ask me to take over.

I do think FSD will be able to take most the heavy lifting out of driving in the next 2-3 years as long as Tesla keep investing. You will have responsibility and probably have to select lanes, OK over taking, help with complex junctions, but it will deal with everything in between. Will be interesting to see how gentle gentle features are added to get to this, or if they just go BANG, here is v11, it comes with a spanner to get rid of your steering wheel.

 

[pgkevet]

I do think FSD will be able to take most the heavy lifting out of driving in the next 2-3 years as long as Tesla keep investing. You will have responsibility and probably have to select lanes, OK over taking, help with complex junctions, but it will deal with everything in between.

Which is a more advanced version of today. BUT the huge risk here is that as driver take-over need is reduced then complacency sets in and if car cannot visually recognise the need for driver take-over - as opposed to gps - then the disasters following would probably lead to a backlash against the tech (as well as whiplash )

 

[Fly.guy]

Here is a copy of a post I put up on SpeakEV a few months ago [before the site was covered in adverts and became unusable] that might be relevant...

Driving to work today, I was tied up in the normal heavy, but slow moving suburban/urban traffic.

All around people driving their cars were voluntarily ceding their 'official' priority on the road to let people out of side roads, give space for people pulling U turns, waiting a bit longer than needed to allow people to cross the road, making space for bicycles coming up the inside and generally being considerate and respectful of other road users. These moves were sometimes indicated by a little wave, or maybe a flash of the lights, but also often simply by a pause or deceleration that allowed a space to open that allowed someone else to slip in.

It all worked and while nobody was getting anywhere fast, we were all making progress with relatively minimal stress.

What I'm having difficulty understanding is how these very human, social interactions, dependent as they are on mutual understanding and knowledge of how these situations 'work', can be replicated by autonomous driving systems.

How will my FSD AI know that someone just gave it a wave and that that specific wave means that it has a very short window in which to nip out in front of someone who formerly has right of way ? How will it distinguish that wave from a wave of thanks or someone having a vigorous hands-free phone conversation ?

It just seems beyond the realms of the feasible to have an AI reading and responding to the subtle social cues that humans make and receive in these situations.

As such, I do have to wonder what will happen when people start to let their AI drive them to work. I can rapidly see a situation developing where an AI, bereft of this social intelligence, fails to respond in a way anticipated by real human drivers around it. As strongly social animals humans are very highly trained to be sensitive to such cues and are also strongly conditioned to punish those that fail to abide by them. I can anticipate that even a small proportion of socially inept AIs on the roads will trigger anger and frustration in the surrounding human population. This risks increased human aggression towards AI cars and further hesitation and paralysis by risk averse AI systems...

I hope I'll be wrong. But I just have some serious doubts that AI-driven autonomous driving is ever going to work in a context where it is having to mix it up with a significant population of human drivers.

 

[pgkevet]

@Fly-guy While I do agree with you there as a comment by Elon on his autonomy day that they expected to be able to interpret pedestrian facial expression - from the viewpoint as to whether they might step into the road. Whether you believe they can do that or not isnlt the point ... at least they are thinking about it.