Lobbyist for tesla/uber gave me the timeline for FSD

[Chaserr]

FSD is a contradiction in terms. Implies Level 5.

Tesla and everyone else are trying to achieve Level 4 Full Self Driving - any road surface, but that's it. Level 5 is anywhere off road - beaches, forests, the moon, doesn't matter - because level 5 is total human driver awareness capability. Uber, Tesla, all of the companies trying to make FSD happen are avoiding level 5 because that's not worth doing yet.

In short level 4 is a decent computer driving on roads with no human help. Level 5 is the turing test for cars.

 

[CHINI]

Let's break it down to three parts.

What you thought you were buying - A package that would allow your car to take over complete responsibly for driving within a couple years. Not sure what you were expecting, but Elon has been claiming two years give or take. Where the first year would be development, and it getting better. Then the second year would be getting regulatory approval.

What you actually bought - A system that would improve it's driving capabilities over time. Like enhanced summons next month, and then city driving NoA in 6-12 months, Where in all likelihood your would be responsible for any mishap that happened over your entire 3-5 ownership.

What your friend probably knows more than me about - The regulatory situation as it exist in California for self-driving cars. Five years sounds plausible to me.

With those three things I tend to only recommend FSD to people extremely excited by FSD, and want to partake in the journey to it.

I do want to say that I'm extremely happy with my purchase and would do it again in a heartbeat. I think Tesla is one of the best car companies in the world and am excited for what the future holds. What I don't like is the misleading statements. Don't say FSD will be available soon, if its more than a couple of years. Might as well add a flying cars option for 10k which will come out in 20 years.

 

[azred]

Tesla has not claimed FSD will be complete by the end of the year, only that it will have the listed features. I expect it will work about as well as AP2 at the start. It might be easier to drive without it than with it. But it will at least try to drive everywhere. Then over the course of a couple of years it will improve to a level that is fairly usable. That's what we've all bought. Heck, they originally sold this final part of FSD for less than EAP. Robotaxi would be great, but not what I've been expecting for $3k. Five to ten years for Robotaxis? I wouldn't be surprised.

I think you are right that FSD on the streets will have similar reliability issues to EAP on freeways. But the stakes are higher. Fatalities to date on autopilot were arguably preventable if the affected drivers had been ready to take control. But if a car blows through a red light on FSD it may be virtually impossible for a driver to take control in time.

 

[pgkevet]

I believe there are a number of flaws in the approach being used. Most importantly there is no evidence of the car making use of distant data and the anticipation that a human driver uses. Easier to understand is that whilst Tesla has masses of data from their cars already deployed, that data overwhelms human analysis and any computer analysis can only look as aspects the programmer requests which by definition misses most of what Elon calls corner cases. Most significantly any analysis of situations where the human driver takes over from FSD is reactive as in after the event as opposed to avoiding the event. Also by the nature of humanity half the drivers will be poorer drivers than the better drivers and therefore generate more data and poorer quality data.
Lastly the vehicles lack several optical inputs (if one is going to depend on optics). There should be side cameras on front and rear bumpers for cross traffic and there should be more forward cameras or a gimble camera for hump-back bridges or dips.
I've also been in the situation where a dog was in front of the car and invisible to the camera cluster and not showing up on front ultrasound - presumably because the fur absorbed rather than reflected. (I knew it was there so no harm done).
Such issues are functionally unacceptable in any claim for even geofenced use full level 3 - otherwise the driver has to pull out from every junction and confirm every pull-away from stationary.
Personal view is that there will be a whole new catalogue of sensors before we get the promised 'meet you at the front door and take you to work' car that Elon reckoned to have 6 years ago. HW3 is out, HW4 is already being developed... you can guess what number we'll be at before this really works 5?

 

[Ofarlig]

I am hoping we get legal to use level 3 on highways in the 5 ish year timeframe, thats about as much as I dare to hope for. If I get that I will be a happy camper though as it will completely change my travel pattern.

 

[mvotb]

FOOL self driving isn't going to happen while any of us own these cars.

 

[cduzz]

I suspect that well prior to "full self driving" we'll see situations where a company like uber manages a fleet of cars without drivers on strictly limited routes only when there are pre-agreed-upon weather conditions. Imagine "contract to take people from this parking lot to this healthcare facility when the sun is out and it is not raining / snowing / foggy." or "take people from this neighborhood to this parking lot downtown using this strictly delimited route."

A company like uber has the resources to send human drivers to "rescue" stranded passengers if the robot car encounters an issue it can't cope with.

Also, it is totally foreseeable that "uber" (robot taxi fleet operator) will go around and buy cars with a set of particular features required to support a required function making those cars more valuable than similar cars without those features.

The timeline for me summoning a car from an airport parking structure in San Diego and having it pick me up in Lake Tahoe in a snow storm... yeah, that's a tall order.

 

[JayyyDeee]

Although I'm firmly in the "three to five years is the dream" camp, I do think that FSD will happen rather suddenly. The gains will come on in an exponential way simply because of the way programming is done now. Will be close then all of a sudden will be there. But, could it be 10 years? Yes, I agree, it could be.

And, there is not unlimited money going towards waymo. Alphabet is not going to send everything it has that way, they do have a budget. Tesla has a fantastic advantage in data. It's collecting it at a phenomenal rate. I think that's the key. I think Tesla may very well be the first to Market.

You literally have no idea what you are talking about. I'm betting almost zero comp sci knowledgeable or even how corporations work

 

[Dithermaster]

Tesla and everyone else are trying to achieve Level 4 Full Self Driving - any road surface, but that's it. Level 5 is anywhere off road - beaches, forests, the moon, doesn't matter - because level 5 is total human driver awareness capability.

That's not correct. L5 is still roads, but all roads. L4 is a subset of roads.

L4: "The car can operate without human input or oversight but only under select conditions defined by factors such as road type or geographic area."

L5: "The driverless car can operate on any road and in any conditions a human driver could negotiate."

Nobody every mentioned forests or beaches. It's still specifically roads we're driving on, not the moon.

See Self-driving car - Wikipedia

 

[Wol747]

Full disclosure: I’ll be receiving my first Tesla MS next month and have paid for the FSD option. (Not that I expect to be around if it ever appears but that’s me.)

In amongst all the hype and hubris from the industry there seems to be a disconnect between safety and usability. Rightly, FSD safety is given the highest priority, but the - rather obvious - fact that even level 4 would be effectively useless except as a gimmick if the safety aspect places constant restrictions on where you can go, is rather glossed over.

FWIW, my own read on this is that REAL FSD - which would only be that if the vehicle didn’t have any manual or override controls - will only come about when/if it has *real* AI, not just a hugely complex set of rules run by millions of lines of code.

And that - AI - would open a different, and huge, can of worms!

 

[BigNick]

For a long time people have underestimated what it will actually take to make self-driving cars, the human brain is amazingly complex, and can in some cases track a hundred or more objects moving in different directions at different speeds. And we only have two forward-facing "optical sensors" a couple inches apart, but they are attached to a nice three-axis swiveling mount.

No one wants a disaster, but a disaster caused by a self-driving vehicle could possibly kill it off for good, at least for our lifetimes. A disaster on the scale of Le Mans 1955 which led to auto racing being banned in Switzerland (and it didn't happen there) for five decades, and that was before the hyper-litigious society we currently live in existed. Worst-case scenario would be a self-driving semi truck (any manufacturer) crushing a school bus and killing all onboard. That would probably be a self-driving-vehicle killer.

True FSD would have to deal with things most of us aren't thinking of here.

Think of the worst driving conditions you've ever had to drive in such as snow coming down at the rate of three to four inches per hour, then suddenly coming up on a multi-vehicle wreck with distracting flashing lights everywhere, and a human directing you to turn around, the road is closed ahead. Or even worse: the pile-up is there, but no first responders have arrived yet. A level 5 vehicle would have to either be able to find a way around, or return to destination, while somehow ensuring the safety of its occupants.

 

[Saghost]

When I purchased my model S recently, I was under the impression that FSD would be coming out in the near future, so I elected to choose it as an option. Yesterday I went to a bbq at my close friends house in menlo park and he is a lobbyist for uber/tesla/waymo autonomous vehicles. He is the one writing the law for it here i California. After small chit chat, i asked him about the timeline for FSD and he chuckled and said "Not for a very long time... Maybe like 5 years. Actually, probably closer to 10 years." He went on and said that waymo will be the first company to go full autonomous and the rest will follow suit shortly after, but that its again 5-10 years away. So now, I'm sitting here wondering if I misinterpreted all the posts I've read/sales person or if everyone already knew this. In 5 years, I will probably have a different Tesla/another car so i'm unsure why I paid 6k for it. I mean, I do like watching it summon, but for 6k?

It wouldn't surprise me at all if Waymo is 5-10 years away from level 4 or 5 autonomous driving using their current approach.

It would surprise me greatly if they were the first to get there.

 

[Chaserr]

That's not correct. L5 is still roads, but all roads. L4 is a subset of roads.
See Wikipedia

Wikipedia is a good place to start learning about something but not a good place to stop learning, you're at the mercy of the simplicity of whoever distilled the article for you from elsewhere or their own ideas.

Level 4 is defined by geographical limitations - this can be only on specific roads, or only on paved roads which is what everyone is doing.

Level 5 has no limitations on geographical position, it simple defined autonomous driving in any human-possible situation.

That's why I pointed it out, even you thought level 5 meant level 4.

The levels are almost pointless, but if we're using them they define the difference between level 4 and level 5 as geographical constraints versus unconstrained. Your wiki does point this out but it's unexplained so you didn't read it as intended. "All driving modes" in your link does not have any limitations on geographical location, unlike level 4 which is typically going to use those limitations to define roads and recognizable road surfaces.

 

[sandpiper]

When I purchased my model S recently, I was under the impression that FSD would be coming out in the near future, so I elected to choose it as an option. Yesterday I went to a bbq at my close friends house in menlo park and he is a lobbyist for uber/tesla/waymo autonomous vehicles. He is the one writing the law for it here i California. After small chit chat, i asked him about the timeline for FSD and he chuckled and said "Not for a very long time... Maybe like 5 years. Actually, probably closer to 10 years." He went on and said that waymo will be the first company to go full autonomous and the rest will follow suit shortly after, but that its again 5-10 years away. So now, I'm sitting here wondering if I misinterpreted all the posts I've read/sales person or if everyone already knew this. In 5 years, I will probably have a different Tesla/another car so i'm unsure why I paid 6k for it. I mean, I do like watching it summon, but for 6k?

Personally, I think even 5-10 is wildly optimistic for FSD. I've been telling people minimum 20 years for true FSD, but that we will have limited FSD before then in gradually increasing sets of circumstances.

In order to implement hands of brain off, sit back and watch TV FSD, you need to have a system that includes redundancy. You'd never want a splatter of mud getting onto the camera, causing the car to get confused. On limited access highways, a lot could be done using high resolution GPS (accuracy on the order of cm) well mapped roads and v2v and v2i communications. Those systems could be used for the basic navigation / path planning work, and then the visual systems could act your redundant channel.

On pedestrian access roads you'd need more. You'd also need a redundant channel to detect whatever 'stuff' finds its way onto the road - people, animals, bicycles, escaped wheels, cardboard boxes, garbage bags, etc... LIDAR maybe, but even that may not be sufficient.

People always talk about the human being only having visual input. That is the case, but processor behind the two cameras has an incredible level of contextual awareness that makes up the minimal level of input, and that no computer will match for a good long while. And let's not forget that that processor screws up pretty regularly as well and people die. Lawyers will not be as forgiving of a computer as they are of human failure.

 

[bob_p]

The goal is to make Tesla's FSD safer than a human driver. Human drivers operate on limited visual and audio input - and have accidents and make mistakes.

While FSD vehicles could add additional sensors to provide great coverage in all areas, minimizing dead spots, coupled with software that could almost completely eliminate accidents - that isn't the current goal for FSD (though that could be a long term goal for future iterations).

Human drivers can see objects in one direction at a time - limited to about 180 degrees and a small viewing area in the side/rear view mirrors. Humans don't have the equivalent of radar or lidar - object recognition (type, size) and relative movement (distance, direction, speed) are estimated using visual input from two eyes spaced closed together.

Using radar and lidar provides more information, and makes it easier for the software developers to implement visual detection - but that doesn't mean software couldn't (eventually) perform the detection without radar or lidar data - it's just harder to do this without only the visual input (even humans have difficulty with this - like the labels on the side mirrors warning that "objects may be closer than they appear").

Like any new technology, there are multiple approaches being used for achieving FSD. All, some or none of these may succeed in the next 5 to 10 years.

Tesla's approach is interesting - relying more on AI and fleet learning and less on more expensive sensors. It might or might not work - same can be said about all of the other approaches, because each strategy has its advantages and disadvantages.

Using HD road maps or relying on V2V/V2I communications can help - but also have flaws - and may not be counted on to achieve FSD.

Haven't seen anything from Tesla or the other manufacturers about using audio sensors to detect objects like trains or approaching emergency vehicles that are not visible.

And what about detecting and interpreting manual signals provided by police or first responders - at sporting events, accidents, construction areas, ...

Musk has claimed FSD will be "feature complete" around the end of the year - what's more likely is that this phase of FSD will be feature complete, handling the inputs they are concerned about now - and will find additional features will be needed to handle the cases they aren't thinking about - and why it takes years for human drivers to operate vehicles safely.

It's much more than lane keeping, maintaining vehicle spacing and following a navigation route...

 

[sandpiper]

I think we're saying more or less the same thing. My main point is that the current attempts have no redundancy and I cannot see anything resembling true FSD without that. And I also think that the "safer than a human" target is fantasy. That will mean exactly nothing when the lawsuits start flying. FSD could kill fewer people, but because it has very different flaws than a human, it will kill a somewhat different set of people. And the lawyers will bite into that really hard.

I think it will have to be MUCH better than a human before we're allowed to sit back and watch TV. Five to 10 times, I would guess.

 

[willw64]

Interesting discussion for sure, and I agree that the notion of real FSD is more like a fantasy, and IMO isn't necessary. I like the passion and the drive towards FSD, whatever that eventually entails, will only make the Tesla driving experience all that much better.

Not to let Musk off the hook, but if I can get very reliable highway FSD where I can essentially be a pilot monitoring the auto pilot, and in city auto pilot with better self parking execution, some sort of advanced summon, and recognition of stop signs and stop lights so I don't have to constantly go in and out of auto pilot I would be a very happy customer.

 

[henderrj]

Slowly more and more people are realizing that FSD anytime soon is a pipe dream of Elon's.

I rented a Model S in December and decided to buy one. Started researching in January thinking buying used between AP1 and AP2+ for the FSD promise. The more I researched and thought about it critically, the less I believed that FSD would happen anytime soon and decided to go for AP1 and save money. I ended up with an AP2.5 car due to the March Q1 sales push (big discounts on inventory cars) but I have no belief that my car will ever be self-driving, and I plan on keeping this thing for 8 years. Thing is I do a LOT of research when making big purchases, I bet most Tesla buyers, especially those who went with FSD did minimal research on their purchase, at least related to the FSD option they're buying.

I think throughout this year and early next year, the illusion that FSD will happen soon will fade away. Tesla knows this, it's why they've re-vamped pricing and removed EAP since everything they can sell is just an EAP feature. By moving that stuff to FSD, they can get a little more money because people will have to choose FSD if they want autopark, lane change, summon, etc. Plus with the new FSD definition, Tesla won't have to put FSD purchases into deferred revenue anymore since they deliver most of what the "FSD" package is now.

They also probably wouldn't be wasting time developing a pickup truck if FSD is anytime soon. FSD doesn't make much sense for pickup trucks, at least not if they're being used for actual work. Parts would be much better spent on building Model 3's and Y's to throw into a robo-taxi fleet if that was feasible anytime soon, but it's not. That's why they're making a pickup. They need to sell cars for everything OTHER than FSD, and that's definitely something they can do as the cars are awesome. They're going to need to somehow think of a way to back away from FSD without tanking the stock.

You literally have no idea what you are talking about. I'm betting almost zero comp sci knowledgeable or even how corporations work

Oh, only 40 years in the business, starting with a fortran college course in the seventies. Nah, not much. And then I owned my own consulting business working with fortune 500 companies for a few decades. But, I'm out of the business now so ...

 

[CHINI]

It wouldn't surprise me at all if Waymo is 5-10 years away from level 4 or 5 autonomous driving using their current approach.

It would surprise me greatly if they were the first to get there.

Why do you think that? Living in the bay area, i see more waymo and cruise vehicles than any other ai car in the downtown areas. Waymo has the deepest pockets while cruise uses lidar. Everyone and their mom have a tesla in the bay, but I can't imagine that tesla will be the first company due to poor data with drivers behind the wheel.

 

[Saghost]

Why do you think that? Living in the bay area, i see more waymo and cruise vehicles than any other ai car in the downtown areas. Waymo has the deepest pockets while cruise uses lidar. Everyone and their mom have a tesla in the bay, but I can't imagine that tesla will be the first company due to poor data with drivers behind the wheel.

Did you watch Tesla's Autonomy Day event?

It was by definition a dog and pony show, but they made a lot of really good points about Tesla's unique advantages in the quest for Autonomy.

Having half a million cars on the road experiencing the challenges while humans drive them gives a breadth of data and difficult events the others don't have - data you need to program a car to deal with those events.