FS_FRA
Member
FWIW Frankfurt up ever so slightly (+0.72%) this morning, albeit on practically no volume.
You can install our site as a web app on your iOS device by utilizing the Add to Home Screen feature in Safari. Please see this thread for more details on this.
Note: This feature may not be available in some browsers.
I'm on record as saying full self-driving won't happen for at *least* five years more, maybe decades -- it's just much harder than people think it is.
Honestly I'm not sure I've ever encountered a truly blind curve.
Right. I have, there are three in my town.
It is simply not useful to make an "average driver" car. Psychologically, people won't accept it: we have documentation that people will demand that an autonomous car be *substantially better* than an average driver; for some reason people are way more unhappy if hurt by a mediocre-competence robot than by a mediocre-competence human. Musk knows this, and has quoted a need to be 10 times better and 100 times better at various times. You will not get this by replicating average. It may reduce crashes if deployed as "enhanced driver assistance" (people will accept it if there's a driver to blame) but it will never be allowed as "full self driving".
Is it not becoming clear that Tesla has made an extremely significant tech breakthrough with this technology? A breakthrough that is probably in itself worth many billions when applied?
I feel this is yet another largely unnoticed moat TSLA has, and is not valued in whatsoever.
I don't think I replied to these FSD skeptical arguments yet, but I think FSD is important to Tesla's valuation and thus to its short term stock price (see my FSD prediction at the end of the comment):
And in support you listed a number of 'hard' scenarios that FSD AutoPilot won't be able to handle, one of them is blind curves/corners:
You seem to agree that Tesla is pretty advanced in the 'vision' part of full self-driving, which creates a 3D scene of objects and velocity vectors. Your argument is that 'driving' in that 3D scene is the hard part and that Tesla 'hasn't even begun' that process yet.
I hope that's a fair summary of your arguments - let me know if it isn't!
I disagree with your arguments:
I.e. I believe the focus of the 'driving' part will simply be physical safety, which consists of the following elements:
- Here's a video, at timestamp 1:34, what AutoPilot V9 sees in a more or less 'blind corner'.
- As you can see AutoPilot refuses to recognize a lane/road that's not visible enough yet.
- I believe the 'driving' logic will simply detect that scenario in your town and declare the road a dead end, stop the car and alert the driver - or if there's no driver alert a remote driver or route around it using another approach.
- Full self-driving doesn't mean the car has to be able to drive through every scenario a human would risk. Full self-driving is likely going to be a very, very paranoid and careful driver who is easy to scare into defensive driving and outright refusal to drive.
FSD driving is the typical 'perfect is the enemy of good' problem. FSD vision, perception and reaction times have to be better than humans, but driving doesn't need to be perfect or even more efficient than human driving, it needs to employ safe driving - and refusing to drive in certain scenarios is an entirely accepted outcome.
- Can the car see the road. If not, it won't drive there.
- Are there any objects (cars, pedestrians, obstacles) on the road that prevent the car from driving there.
- If all criteria are met, drive.
You also seem to not take regional differences into account: road systems in the U.S. are much simpler, larger and generally safer (in terms of layout, not necessarily in terms of drivers) than those in Europe. If Tesla is able to tap FSD for the U.S. market only that's a huge chunk of revenue already.
Finally, you also don't seem to take customer differences into account: FSD will be most disruptive to the transportation/logistics business. But trucks are already limited in where they can drive - and it's quite probably that an 18-wheeler won't typically go on routes with blind corners, right? They'll stick to highways and easy to access logistics centers.
Just because FSD will initially have trouble mastering narrow Irish, Welsh or Scottish roads and towns without disengaging doesn't mean the valuation of Tesla's FSD business has to be zero.
I believe here you are conflating 'safe' driving with 'efficient' driving, which are two different attributes.
FSD needs to be an order of magnitude safer than human drivers - and it's probably already at that stage or even better for highway driving, and AutoPilot is probably already a two orders of magnitude safer driver in bad weather (due to the radar that sees through rain and fog) and at night (due to 8 cameras that are sensitive into the infrared), and of course due to being 100% alert all the time and never tiring.
FSD doesn't have to be efficient at driving to be useful, at least initially. It's quite likely that human drivers will be able to master tricky driving scenarios a lot better than FSD software for many years to come. But the weird driving scenarios are not what matter: what matters are the 1 billion vehicle miles driven in the U.S. every year, and whether FSD can handle a meaningful percentage of those routes. It will be self-correcting after that: customers will self-select and those will adapt FSD faster who have higher utility for it.
TL;DR: As a contrast to your prediction 'of at least 5 years but possibly decades', I do think Tesla is going to release the first usable versions of FSD next year already: traffic light recognition, road sign recognition and maybe driving around corners that have working traffic lights, helped by GPS and maps.
It will still be a 'paranoid' FSD that falls back to the human driver frequently, but it will be able to drive autonomously in a surprisingly large percentage of the real routes that people take, and this percentage will increase with every new release in a largely organic, iterative process. I also expect there to be a real San Francisco to New York FSD road trip attempt by Tesla next year, with a very low number of disengagements.
Gave you a funny because here in Germany speed differences often are *much* greater. But I (and scientists looking at efficiency of traffic flow and safety) agree that it is better to keep them low - and much more relaxed if when passing, you don´t have to expect someone coming from behind at 50mph faster than you.
Yes, I think they are ahead. In tech a lead does not last long. Others can soon emulate the board and the unified two frame input.
What they can't copy is the number of hardware equipped cars on the road. Nor the volume of learning data Tesla has available. Nor the fact that Tesla's cars are electric with low low running cost - no ICE can compete in a no paid driver scenario.
In other words, if the battery production is currently the limitation factor, MR can allow to make ~20% more cars with margin 13.2% (instead of 15% that the LR has).
Looks like Panasonic incurred additional opex because they had to ramp battery production at Gigafactory 1 faster than initially anticipated.
I’ve been curious about what happens on the Tesla network when deteriorating weather conditions force the fsd car to simply pull over and drop driving. Would tesla require a licensed and capable driver to take over?
BTW., this should be a reply to those people who asked "why didn't Panasonic ramp up faster??". The reason is that it's very expensive to expand cell manufacturing, and Panasonic has to front these investment costs. (Tesla leases it from Panasonic and finally ends up owning the lines, but that takes years of lease payments.)
So Panasonic had to be conservative in their ramp up: they only committed to expanding cell capacity one Tesla demonstrated a Model 3 week with 5,000 vehicles produced.
It's all good now: Panasonic will be sharing the cash Tesla makes on all these cells in the quarters and years to come.
TL;DR: As a contrast to your prediction 'of at least 5 years but possibly decades', I do think Tesla is going to release the first usable versions of FSD next year already: traffic light recognition, road sign recognition and maybe driving around corners that have working traffic lights, helped by GPS and maps.
Tha
Thanks for the update. Can you please advise the lease terms between Panasonic and Tesla?
I haven't heard of tesla making vans. Where did this rumor start?
I'd think it's way low on the priority list, way below Y, semi, and pick up truck. I guess Amazon does use those Sprinter vans...and some local chain grocery stores use those little commercial vans so there is some demand. I worked for a new car dealer who carried those commercial van models and we hardly sold any of them, ever. I don't think I even witnessed one sold over a 4 month period. The model Y will be a game changer though. Crossovers were easy sales.
Your argument basically adds up to: "current Tesla tech + Tesla's NN chip early next year + shadow mode on growing amount of Tesla vehicles = Neural Net can figure FSD out exponentially fast."
Bonus: What's in a name. I basically disagree with your definition of FSD.
Elon Musk himself: Elon Musk says Tesla likely to build an all-electric cargo van on its… (link to an Electrek article on archive.is (Fred doesn't get the clicks))
I see tons of Ford Transits (and Transit Connects, but that's something smaller), and I used to see a fair amount of Mercedes Sprinters (although the Transit being cheaper displaced them). Note that the vans mostly get sold through fleet sales, so things may look very different for how they're sold.
And, it'll likely be a simple variant of the pickup, so it could probably be released alongside it.
- S&P 500 inclusion requires four profitable quarters in a row by some sources, five by others, i.e. it would happen around this time next year or early 2020. Current inclusion criteria used by the S&P 500 committee seem to be somewhat opaque, but I don't think there's a rule to have a profitable year, is there?
Eventually, once the legal/regulatory environment permits it, there would be "remote drivers" which would basically be a service you could hire from Tesla: the remote driver takes over if there's some trouble, …
… and there's also a network of FSD-service cars that physically drives out to any permanently stuck FSD car. This would be similar to roadside assistance services, a bit higher cost because FSD troubles are expected to be much more frequent initially than mechanical failure. Pricing might initially be also 'per incident' or 'per minutes driven', not flat.