I’ve thought the difficulty was being vastly underestimated for a while. A reusable Lunar and Mars shuttle system is probably a lot easier than actual FSD.
I worked full time in a computer vision company (non-automotive) for several years prior to 2018. I joined an automotive company (not Tesla) in 2018 and very intentionally steered clear of the autonomous efforts going on there. Not because my skillset wasn’t best applied there, but because I didn’t believe the roadmaps the industry was laying out at the time. “Level 3 by next year, Level 4 by 2020, Level 5 by 2022” is what they were saying back then. I called bullshit (silently) and worked on viable software instead.
Lots of people were doing the same. This isn’t surprising to a whole bunch of people that work in the non-automotive computer vision/ML space and realize how absurd it is to think the these models will converge any time soon.
We're not too far behind on that estimate. My guess is in less than 3 years - Waymo will be publicly available in large parts of Phoenix and probably another major metro - and within 10 years most of the south west.
But a lot of uninformed people thought we might have FSD everywhere by now. I think there's just less uninformed people now.
If they change everything about their approach, and/or lease tech from Cruise or Waymo - then maybe they'll have something resembling FSD in large parts of the southwest. Predicting 5 years into the future is tough - 15 years is almost impossible.
Anyway, the pendulum seems to have swung the other way now. Since FSD is so far behind Elon the marketing-hype-stock-pump-Charlatan / the louder voice in the room - now the mainstream idea is - we're never going to have FSD.
This is ironic, since we basically already have it in a few large metros. It seems like the newspapers famously publishing that we might have manned space flight within 10,000,000 years - literally the day before the Wright brothers first flight...
People thought the car would never catch on, because cities were designed for people and bikes and horses. The car was so useful, we redesigned cities. Self-driving will be the same way. Once it takes off in some places and the value is realized, cities will redesign to make it work more broadly. It might take a long time, but it will happen...
Unless the nature of transit changes and individual vehicles are phased out for whatever reason first - though I don't see that happening without FSD busses / street cars.
The latter model doesn’t scale, and can’t handle surprises.
Or who knows, maybe it could scale if you had a global fleet of specialized vehicles and gigabit/s full-duplex internet connections in all cars. But that’s a different approach even to Waymo’s.
No no, only musk was stupid enough to imply level 5 autonomy within x years.
Everyone know how hard it is, realtime accurate position, machine perception of roads, & accurate prediction of cars/lorries/other are all very difficult problems. Then to hobble your team by saying "blah blah blah no lidars" because they are expensive and someone from the ML team convinced you that they could estimate depth reliably, was the clincher.
Level 4/5 self driving is more of a matter of scale. if you have a 99.5% safe rating, then a lot of people over a country the size of the states are going to have a bad day.
Tesla had to use all their weapons including fraud, exagerations and lies because their core business is the worst business in the world: manufacturing automobiles.
So they have essentially nothing to fall back on if FSD fails.
I'm sure there really were true believers who got sucked in by the rapid advances in ML and figured "How hard can the last 20% be really?"
You had companies that figured they'd better their bases because they could afford to and FOMO.
You almost certainly had some number of cynical grifters pitching a story for their investors that they knew was nonsense.
And you had a lot of people who saw all of the above and figured that it must be all true.
Assuming stupidity is a bit over the top.
Promising that it be ready by 2018(or whenever it is) == unrealistic ambition
Chucking it out the door, branding it as FSD, charging for it, and not actually putting in the right sensor suite to make it work == stupid, or, perhaps dishonest.
But yeah, at some point he should have wised up and stopped saying that FSD was gonna be ready Any Day Now.
It’s the worst sort of problem, one where 10% of the work gets you 90% of the way there and the other 90% of the work is needed to get you the other 10%. Problems like this are really bad for creating this mirage that you’ve almost got it when in reality you are far from the end.
The former has value and people will buy it even if the latter can't be delivered in totality. Making Tesla rename their product does not mean their cars magically stop driving themselves, and especially does not mean that there isn't a market for people (that's me, hi!) who want cars that can drive themselves.
It doesn't leave space for interpretation in which only specific parts of driving are possible. At least in my mind.
But that's not regulatory approval, and coverage of the issue tends to conflate the two. So you have people like you who read this stuff, but don't get rides from friends, who think somehow it's a disaster and a failure. But it's an actual product people are buying and using. Add that to all the folks who have a vested interest[1] in Tesla failing, and... the discourse around this subject is pretty toxic.
All I can say is, again, call a friend and get a ride. There's a lot of hate to push through, but the cars are amazing.
[1] Because, to be fair, the CEO is an asshole burning Twitter to the ground as we write.
So... I'm actually fine with renaming it. People shouting "fraud" on the internet want somehow for a different name to magically make the product bad. But the product is great.
[1] Yes, including Waymo and Cruise, who work only on carefully calibrated geofenced areas. My Tesla will take the highway when it needs to and honor a five-way stop sign. Waymo won't even make an unprotected left turn.
The most important metric for self driving is the distance a car can travel before it needs assistance from an occupant. That’s what self driving is. Tesla is very far behind the leaders here.
Also, Tesla might not in theory be geofenced, but in practice it will refuse to drive on many roads making it effectively geofenced.
That's reductionist to the point of meaninglessness. A car limited to a specific track could keep going autonomously until it breaks down, thus having an infinite metric on the distance a car can travel before it needs assistance from an occupant, yet practically it could easily be useless as a self-driving vehicle.
Being able to handle self-driving without being restricted to carefully chosen "good spots" and instead being able to generalize is at least on par in importance with distance the vehicle can travel.
Also, the vehicle determining that it can't handle a certain road well enough yet is a desirable feature and isn't really comparable to a bunch of people deciding if a certain road is safe for their vehicle before allowing it to self-drive there, as the latter simply isn't practical at scale.
There are under 4.1 million miles of roads in the US. ~70 people could map every mile of that every year assuming average speed of 30mph and 2,000 hours of work per year. It’s literally trivial compared to building a self driving system.
Tesla’s insistence on good road markings is a far larger limitation than simply pre mapping roads.
One of the problems with this discourse is the propensity for people to talk past each other. No one is interested in mining automation here, no matter what level of SAE autonomy they qualify for.
> Tesla’s insistence on good road markings is a far larger limitation than simply pre mapping roads.
And another problem is people's insistence on arguing from bad information and ignoring the testimony of those who have good info. I honestly don't know what you're referencing here, FSD works fine on unpainted roads (I live on one).
Yet, it clearly disproved your point. A self driving Taxi or Semi needs to be fully autonomous to be useful, coverage area is a useful feature only after you can get rid of the driver.
I’ve seen a FSD Tesla have issues with bad road markings recently. That’s a complete failure from an autonomy standpoint, because if they can’t predict and thus avoid bad situations the system is completely dependent on a human driver.
And that’s the problem, Tesla has been stuck at at the cool tech demo stage for years and adding features alone isn’t progress because something that generally works as long as a human can take over is so vastly easier they can just gloss over the hard bits. Eventually they will run out of new feature that sort of work and the cars will still still be unable drive anywhere by it’s self. Perhaps they start trying to level 4+ in a few years, but their going to need to start over getting each feature to actually work without that safety net.
https://www.caranddriver.com/news/a41117967/ford-bluecruise-...