Tesla FSD data is getting worse, according to beta tester self-reports
electrek.co
electrek.co
- Only allowed on limited-access divided highways with no stoplights, roundabouts, or other traffic control systems
- Limited to a top speed of less than 40 mph
- Operates only during the daytime and in clear weather
But the big promise from Mercedes is that it would take legal liability for any accidents that occurs during Drive Pilot's operation, something that Tesla doesn't appear to be even thinking about wrt Autopilot and FSD.
I would love someone to goad/challenge Tesla to step up to Mercedes. If FSD is so much better than Drive Pilot, then why doesn't Tesla agree to provide a "safe mode" for FSD, that operates with the exact same restrictions as Mercedes' D-P, and offers the same legal protections to any users who happen to get into accidents during "safe mode" FSD operation?
[0] https://www.roadandtrack.com/news/a39481699/what-happens-if-...
Mercedes has these limitations not because their tech is less capable than Tesla's but because Mercedes is a real car company with real engineers and a gold-standard reputation to maintain.
Tesla, in contrast, is a software company that is trying to take "move fast and break things" into the two-ton 75mph vehicle space, with predictable results.
Mercedes explains the purpose in their press release:
> Conditionally automated driving on suitable motorway sections where traffic density is high
> During the conditionally automated journey, DRIVE PILOT allows the driver to take their mind off the traffic and focus on certain secondary activities, be it communicating with colleagues via In-Car Office, surfing the internet or relaxing while watching a film. In DRIVE PILOT mode, applications can be enabled on the vehicle's integrated central display that are otherwise blocked while driving.
https://group-media.mercedes-benz.com/marsMediaSite/en/insta...
Based on that I'm assuming it's following other vehicles rather than following the road, which does make me wonder what happens when you get a clump of DRIVE PILOT vehicles which are all trying to follow each other.
Edit: It seems like it needs a vehicle ahead and under 60kph and uses a mix of other vehicles and road markings. It seems quite usable (but I still have to wonder how many vehicles you can get to follow each other in a chain): https://www.youtube.com/watch?v=0yiPaKfKLZs
Of course if you prefer, move fast and brake... maybe
IMO that’s just good product strategy.
Mercedes' solution is a car company taking actual responsibility for their software. If they feel the lawsuits/insurance claims/legal snafus are worth the risk, that means their software is probably pretty damn good in that limited scope. Otherwise they could literally bankrupt the company with lawsuits! That's a lot more confidence inspiring to me than Elon's repeated pie-in-the-sky claims.
If I could have read my emails in that time it would have been really nice.
Mercedes is selling a product that has a small set of well-defined cases where it can actually be used.
Being cocky and funny without delivering great results is simply embarrassing.
I think Mercedes and other automakers have a good chance to bypass Tesla now since Elno is captivated by his Twitter acquisition.
I’ve been disappointed in Elon for some time and feel society may have already taken him for all of his good ideas.
So I’d bet the opposite. This is a chance for Tesla to recover and match the established auto makers. And the only chance they’ll get.
This is both pre-sale, where you have to build a lean, mean, fast pipeline from vendors to assembly, and post-sales, where you have to have service infrastructure that spans continents, if not the world.
Tesla so far shows very little signs of being able to do either. And just like Whitley failed and ended up being bought by Mumbai, my personal bet (caveat lector!) is that Tesla-the-brand might survive, but Tesla-the-car-manufacturer would end up a subsidiary of a Chinese car manufacturer, who has the car manufacturing chops, but can't build a brand.
Seems pretty major to me.
But people absolutely will not buy Tesla's because of Elon.
Brands work hard to avoid negative sentiment for a reason.
The worst thing about it isn't the panel gaps or reliability (haven't had any problems). The worst part is Elon Musk and his fans. Shortly after getting the car three and a half years ago, I was leaving an outdoor party and a man who was a Musk superfan was doing that waving of arms of worship that you sometimes see fans do at metal conferences, and it was just embarrassing. Previously the same man had been gushing about full self driving, and I said there was no chance it would be delivered on time, if ever, and he professed his undying trust in Musk.
Combined with Musk's recent anti-Ukraine efforts, his hyper-partisan paranoia thay he's trying to push on Twitter, his hate for trans people, his hate for biological science exhibited throughout the pandemic and even today in his "jokes" about prosecuting Fauci, Musk is waging cultural war against every single aspect of my identity.
I hate Musk so so so much, and I know he had almost nothing to do with the creation of the car I like, but it still pains me everytime I get in it to know that I helped such a despicable person make a ton of money. Never again will I buy a Tesla, especially since there are now competitors. I'm sure I would hate all the rest of the auto execs almost as much if I knew as much about them as I know about Musk, but the nice thing is that I dont know a damn thing about Stellaris's CEO, from their name to their former partners. And there's a lot of value to that, as a customer.
Maybe Musk is just trying to win over conservatives and jackasses to but Teslas, but I doubt it. I think he's just a dangerous fool.
OTOH, people (at least online) like to talk out both sides of their mouth re" Tesla, saying on one hand "Musk isn't an engineer, didn't found the company, doesn't add anything to Tesla, etc etc" and then also "i'll never buy a tesla because of Musk".. like, if those are both the case then what about every other company and CEO? eBay's CEO collaborated to harass a random couple who posted critical articles, do they not use eBay? Adidas made untold billions from collabs with Kanye, do they boycott Adidas? CVS, Walgreens, and Walmart agreed to pay $13bn in a settlement over the opioid epidemic -- literally killing people! -- do they shop at CVS, Walgreens, Walmart, or buy prescriptions from the companies that supply the opioids? Unethical Apple manufacturing practices... etc etc etc.
The point here, which is not 'whataboutism', is that there is seemingly no generalizable principle at play here, that for some reason Musk is a bridge too far in comparison with other companies that arguably do much, much more harm.
Whereas Musk is actually serving as a front man, greedily taking credit in a way that I never even saw Steve Jobs take credit for the work of others.
Musk's actions are also a very personal attack against my loved ones, in a way far more personal than drug stores selling opioids is. It's personal insults, and actions that are supporting the displacement of millions of family members's countrymen, deaths of more than a hundred thousand, and so many children's lives scarred. Pretty hard to compare Musk's actions to what Walgreens management did for the opiod crisis. Personal insults also go a really long ways towards negative polarization, whereas merely distributing prescribed drugs makes it a lot harder to fault drug stores.
Don't discount the impact of the opioid crisis with more than a million dead (https://www.npr.org/2021/12/30/1069062738/more-than-a-millio...) -- those pharmacies did the math and decided double-digit billions was cheaper than a trial -- but I do see what you mean about the personal connection. What I'm wrapping my head around is I don't see anyone (not that that means they don't exist) saying their cousin died from opioid overdose so they're never shopping at CVS again, you know? The venn diagram of people who own an iphone and people who think sweatshops are evil is basically a circle, but Musk tweets something moronic about Fauci and everyone loses their shit.
But I purchased my Model 3 in 2021 because, after cross-shopping many car models and trying to optimize for the characteristics and features I wanted, it was the best thing at its price-point. And even now, even after the Twitter debacle, I still enjoy driving it every day. I couldn’t care less about the man who made the car; the product won on its merits.
However Tesla makes about $9000 in profit per car. If that is cut in half it still beats Toyota’s $1200 by a wide margin.
When people make claims like this its just really fucking baffling. Like, are you so blinded by hate that you are unable to look up basic stastics?
Tesla is growing like gangbusters. In 2023 they will very sell pretty close to the same amount of as BMW, a company that has existed for 100 years.
They are overtaking more and more car companies all the time. They are in range of companies like Geely for example.
In fact, partly because of Tesla car companies have been merging. And these large mega companies are using sales volume.
Tesla is already making more profit then those companies, while also growing faster.
Like seriously, what about making record profits, fast growth and overtaking more and more large car companies in volume says 'incapable of stepping up to be a major car manufacturer'.
> is that it's not about the machine, but about the infrastructure.
> This is both pre-sale, where you have to build a lean, mean, fast pipeline from vendors to assembly, and post-sales, where you have to have service infrastructure that spans continents, if not the world.
You are joking right? Tesla has invested more in infrastructure then the other car makers. The super charger network. They have their own huge direct-to-consumer sales team. They have distribution centers all over the world that they own. They have vertically integrated their service and this for a long time was losing money, but as they grow is increasingly gone make money.
In addition to that they have built huge, vertically integrated factory parks that are bigger then almost everything anybody else is building. Go to Austin and see for yourself.
Tesla is investing in its own battery materials production and making its own battery in the same factory where they are making the cars. Literally nobody else has anything like that.
You statement is literally the opposite of correct. Tesla has exactly done what you suggest, a huge global investment in infrastructure.
> Tesla so far shows very little signs of being able to do either.
I'm sorry but are blind? Have you not look at Tesla numbers and statistics since 2014?
Because what you are saying is straight up delusional. Its not even open for interpretation.
Tesla made profit of 3 billion $ in Q3. That is while growing the amount of Super charger, amount of Service centers, amount of distribution centers, scaling two gigantic factory projects, investing in its own battery production and investing in self driving (including its own data centers).
> subsidiary of a Chinese car manufacturer
The largest manufactures in China are Geely and SAIC.
Tesla will produce more vehicles then Geely by next year and Tesla is already making far more money then Geely. SAIC is still a little bigger then that but Tesla is already beating them in terms of profit.
Are you simply not looking at these numbers?
In the EV market, they #1 and #3 for most cars sold per model. What better results would you expect?
Not only that, but they have a quality reputation to uphold with their domestic (German) and regional (European) market.
Your average discerning German car buyer (i.e. the sort who has a Porsche 911, or a higher-end Audi/BMW/Merc) in their garage will swiftly tell you about numerous problems with the Tesla before they've even sat in it.
Panel gaps, for example. They mean a lot to your average discerning German, and your average Tesla has them by the bucket load.
In fact, the German in-joke is that the reason Tesla built a factory in Germany is so that the Germans could (try to) teach them how to fix the panel gaps. :-)
Speaking more seriously on that front, there's been the acquihire of Grohmann Engineering.
Not a Tesla owner or fan but I have a question that's really bugging me now: Which customers really care about panel gaps?
Do average joes, SUV driving soccer moms or suburban white collar workers go around with a ruler measuring their panel gaps with others like "yeah your Tesla is cool and all, but sorry, my Audi A5 has much tighter panel gaps which is what matters most you know"?
When has the panel gap became "the benchmark" indicative of car quality beyond the body shell?
Like, if the panel gaps are the only thing you can find wrong in a car, then it must be a really really good car, right?
Is there any proven evidence that the panel gaps corelate to quality and reliability of the rest of the car, or is it just a myth of the car enthusiasts community that got spread around and went too far? I get it, some Tesla are unreliable and built poorly, but it's not because they have big panel gaps. The reverse is also true for many cars, so this isn't a rule.
Sure, if you want to measure and compare panel gaps, then by all means go ahead and measure panel gaps, but please don't pretend they mean anything more than that, and that it's somehow an indicative for the car's overall quality and reliability, because so far there hasn't been any proof of this correlation.
The average joe buys a car because of the value it gets them (because every dollar matters)
Does having tighter panel gaps help with the resale value for the average joes?
Notwithstanding the fact that the market for super cars is nothing like the market for Teslas or 5ers.
But my point stands, that panel gap as a benchmark alone is no measure of quality or any other metric.
The reason that there are so many super-expensive pickup trucks on the road is not because people are hauling around things that require a pickup, for example. And when combined with the face that pickup beds are becoming increasingly useless...
To me, panel gaps are a proxy for how much faith you have in your consistency and quality control.
I beg to differ. Modern German cars might have panel gaps tighter than a nun's fanny, but their reliability, especially after warranty is over, is so awful than in no way can I say that they represent quality. Those quality cars went away in the late '80s early '90s when the engineers got replaced by the bean counters and cars became white goods with many critical parts outsourced to the lowest bidding contractor, that must look 'cool' in the showroom, but fail the second the warranty runs out, or many times even before that.
To me, the panel gaps are a superficial metric of quality and prove nothing of substance that goes beyond body shell.
Why don't we measure quality by how reliable a car is over time and how long it lasts? Surely that would prove good consistency and quality control on the manufacturer's side, no?
Tight panel gaps only shows how much efort the manufacturer has put in the body, but says nothing about the quality and reliability of the electronics and mechanics, which is what really matters in a car for most people.
https://www.google.com/amp/s/www.newsweek.com/tesla-ranked-n...
According to the article:
> "the Audi E-Tron and Volkswagen ID.4 were singled out as being unreliable"
So if Audi and VW are also unreliable then the panel gaps prove as a poor signal for reliability, which was my original point.
Edit after your reply below: Sure, Tesla has poor reliability, but not because it has poor panel gaps. Those two can be completely disconnected. Just because they coincide sometimes, doesn't make this a rule of thumb like some car snobs try to convince you of.
You can easily have cars with great panel gaps that are incredibly unreliable, and vice versa. Panel gaps mean nothing more than panel gaps.
As are different models within a single manufacturer. I love my Ford, would highly recommend it, but would never buy or recommend a Pinto or a Bronco II.
The point people typically make is that Tesla has uncommonly poor panel gaps, which point to poor quality and tolerance control in their manufacturing. This is a complex skill that automakers have been refining for 100 years. It is indicative of something, just as the quality of paint job indicates the care and quality with which a hot rod was built.
For me at least, i don't actually think of something like panel gaps as a reliability indicator, but rather it is one of many indicators of whether more quality of life stuff is going to go wrong (ie door rattling, buttons becoming loose or fiddly, etc). These things don't stop me from getting from A to B, but at that price point they're still important considerations.
If a lambo was barreling towards me I could probably just hop over. Not that they would want to go particularly fast anyway in the city, if they care about the underside of the car.
I don't know if I could find a source, but it would probably be something like Forbes in the '80's: Lee Iacocca (of Chrysler) said that they learned from Japanese collaboration that if they made their gaps wider, then they appeared neater because they appeared parallel. And it was cheaper to make a wider gap, and it was easier for QA to pass the cars. Otherwise the cars would have to be whacked with mallets until the narrow gaps looked right, and that slows the line, and that indirectly adds cost.
[This doesn't mean that the Japanese collaborators were making huge panel gaps on their own cars, it simply indicates that Iacocca('s people) got a certain idea from them]
The cars looks like they were in an accident and had their body panels poorly repaired/reassembled.
In a way though you're right - no one cared about panel gaps until Tesla came along because until then even the cheapest of the cheap manufacturers were able to get that right.
Not sure how much this matters as Tesla’s sales are really high, but I think this basic stuff is important.
Average Joes and soccer moms in a typical suburban area cannot even afford Teslas and are extremely happy with their Odysseys, Seinnas and Pacificas. Even an Audi A5 is cheaper than a Model 3 in most cases and Audi's ride quality and cabin noise is extremely better than any Tesla I have ridden in. Audi interiors (though not as good as its other German competitors) beat Tesla by a mile.
Thing is, no one else has them. It's a real achievement for Tesla that the doors don't align with the body when every other car, no matter how cheap, manages to not do that.
What's sold in Europe at least.
So you'll forgive me that I don't trust the rest of the car either.
Source: I've driven a friend's brand new Tesla. The rear right door didn't align with the body.
Also I couldn't figure out how to manipulate the a/c, the turn signal stalk was too smart for me and kept turning the signals off at the wrong time * etc. Too bad about that engine.
* I have a feeling their designers only ever drove on wide american streets that have only 90 degree intersections. For the life of me i couldn't get the turn signal to stay when the main road was going right and i was trying to get on a secondary road that was like 30 degrees to the left. Probably because i was basically going straight ahead from the point of view of the car at the start.
Not once in my life has anyone talked about car panel gaps except on web forums when demonstrating Teslas or rubbish. I had one poster explain to me what is and what is not a luxury car. Alright man, good for you!
Honestly, Who really cares? Perhaps I haven't met the discerning average German car buyer.
You know what bugs me about my car? The fact that the range has dropped by 20% and Tesla support says there is no problem.
As a German I can certainly say that folks here really pay attention to where cars are manufactured. Tesla Model Y's available on various platforms are boldly advertised as "Manufactured in Grünheide". At the same time people are aware that some domestic models (such as the Mercedes EQS SUV) are solely manufactured in the US and then shipped over, which are sometimes conceived as lower quality as a result.
On the Model 3 subrreddit there are endless post about reliability including ones I never thought would be an issue.
Some recent ones include driver side mirror housing falling off and snowing entering the trunk compartment.
Maybe on a Yugo but nothing that purports to be anything better than a cheap as possible econobox.
[0] https://www.speakev.com/threads/wing-mirror-glass-fell-off.1...
Do you have any evidence to back up these claims? On what trends are you basing that on?
Do you have evidence that Tesla performs worse now because of Twitter?
Literally do you have any real world bases for this or is that just what you hope will happen?
I just replaced my Audi Q5 with a Tesla Model Y (which... wasn't $100k). No panel gaps, no other problems, and the overall quality feels nicer than the Audi. Shrug!
Anyway, yeah, the next few years look really exciting for consumer EVs. So many announcements in 2022.
Although one feature I really liked was that it would tell me what the biggest consumers of power was. I would definitely like to have a car with a heated steering wheel in the winter here, but that isn't in the cards.
First thing I did with this Tesla is turn off the heated steering wheel for my driver profile. My hands sweat easily, so it was undesirable.
Nevermind FSD accidents, I think I'd get myself killed just trying to turn down the heat on that giant touchscreen.
The biggest thing that Tesla got right besides the acceleration is the value-add features like Sentry mode, the tight integration with the phone, the things like Walk-away locking (although I would very much prefer it lock closer to my car because it's about 40-60 ft away when it locks and it makes me nervous).
The build quality is cheaper, the sound system sucks, and I generally despise how many things are tied to the screen. I want buttons and knobs for the air system, to have to hunt for that on the screen is very dangerous, I hate it.
The Porsche feels more luxurious but it's mindblowing how they get so many things wrong. The dashboard is much too complicated with a lot of redundant buttons. Something as simple as there's no place to put my phone, there's only extremely awkward locations that cause my phone to fly around the car when I go around any turn, which is so dumb in 2022. The backup camera has way too narrow field of view, I've almost backed into 2 cars in parking lots, something the Tesla got right. At least they have Apple Carplay, but activating it is extremely annoying.
I also have BMWs before (X3 and X6) and overall my favorite car of all time is the X6.
Which isn’t to say you’re wrong, a driving experience isn’t the same, but I’m sure that experience was the guiding principle behind Tesla going full touchscreen.
That being said apple put a lot of work on the keyboard experience with haptics and fuzzy hit design to make it as functional as it is. I’m not sure Tesla has done the same level of usability work.
Trying to find out what mode to be in to turn up the heat when going 70mph is a recipe for disaster no matter how much money they pour into it.
I think it's actually quite worse in terms of security, as you can use some kind of range extender to pretend that the key is close and open (or even start?) the car.
Having owned model Ys made in both China and Germany, I can tell you that the germans have not succeeded. The made in China Y was 100% no issues, while the German one looks like it has been in a collision already. Anecdotal, but this seems to be the pattern with cars delivered to Europe.
I've got plenty of descriptive words for the type of person you described, "discerning" isn't one of them.
Totally agree with all your points against Tesla, but “gold-standard reputation” for Mercedes? Based on what? They are consistently rated as one of the worse brands reliability wise (Tesla usually being worse, but still).
Because they have a reputation of breaking down a lot, and a "cheap" Mercedes is still a Mercedes which is fixed using Mercedes-priced parts.
If you are in that market, and stick with the older models that are proven to be reliable, they are pretty safe buys. Like any used car, a lot depends on the care given by the prior owner, but people who buy Mercedes cars new tend to have at least above-average income, and can afford to maintain them properly.
That said: the newer generation is not for me, I had a C-class with automatic emergency breaking try to murder me twice and I've sold it and gotten something much older without that kind of accessory.
You can have hundreds of qualified people who are called "engineers" and would be excellent in another environment, but if the culture is "If I say jump, you ask how high" you don't have an engineering culture, you have an autocracy.
A car is never going to prevent a determined individual from doing stupid things. But it is a big problem that people who are trying to be responsible are misled about what Tesla's "Full Self Driving" can actually deliver.
And Mercedes simply does not have the ability to produce something like FSD Beta from Tesla no matter how 'real' their engineers are.
Roads need to be premapped ahead of time with LiDAR
Roads need to be pre-approved
Car cannot go above 37 MPH
limited-access divided highways with no stoplights
no roundabouts
no traffic control systems whatsoever
no construction zones
only operate during daytime
Reasonably clear weather
Without overhead obstructions
Let me know of any other system that is even remotely close to being able to do the following:
Without the restriction that hands are on-wheel and driver is paying attention to the road. That's a BIG difference.
> Mercedez-Benz has announced approval of their “Drive Pilot” system, in Germany, which does fully autonomous operation in highway traffic jam situations.
> ...
> The Mercedes car provides the traffic jam assist function — only on German motorways to start — below 60 km/h. While people debate whether they want to drive their car or not, nobody likes driving in a traffic jam, and everybody hates the time wasted in them. As a luxury feature, this will let drivers make more productive use of that time.
https://www.forbes.com/sites/bradtempleton/2021/12/13/merced...
Of course, other driver assistance systems might be close to on par with it, but a system that successfully navigates stop and go traffic 99% of the time is very different from a system that successfully navigates stop and go traffic 100% of the time, in terms of driver attention required.
And yes, I understand that drivers can get lazy with a level 2 system. But if the selling point of Mercedes is taking over liability from the driver, I am mostly concerned how the system would benefit me as a driver and I regularly use my car's level 2 features while paying attention.
Taking liability is an interesting PR move, but I don't think it matters in stop and go traffic where speeds are relatively low and damage is typically minimal if any.
If you've been playing an immersive game on your Steam Deck (or a VR headset), yeah good luck good with knowing what to do when the car bing/bongs you.
If you think people shouldn't play games, what exactly does "you can NOT pay attention" mean? If the difference is solely liability - that's great but it doesn't make a difference to people in other vehicles (or to pedestrians).
Mercedes implementation takes the legal liability. It will definitely stay in its own lane without any driver input. It will continue going again after a full stop all on its own.
And do those adaptive cruise control/lane keeping systems allow the driver to take their hands off the wheel and stop paying attention to the road?
So does the Texas Highway Patrol ticket everyone in a traffic jam? They almost certainly don't, which should be a big clue that it's not so simple.
This appears to be the actual law in Texas:
https://texas.public.law/statutes/tex._transp._code_section_...
> (a) An operator may not drive so slowly as to impede the normal and reasonable movement of traffic, except when reduced speed is necessary for safe operation or in compliance with law.
> ...
> (c) If appropriate signs are erected giving notice of a minimum speed limit adopted under this section, an operator may not drive a vehicle more slowly than that limit except as necessary for safe operation or in compliance with law.
It's not safely operating a vehicle to go 40mph in 10mph traffic.
You don't live in Austin or Houston, do you? :-)
I've been at a dead stop on many highways in Texas, along with hundreds of other cars around me. I see such things happening pretty often outside my office window.
Honestly, times when I'm going <37MPH on a controlled access highway is some of the most annoying driving that I'd like to have completely automated. That usually means I'm in stop and go traffic, some of the most grating time to drive. Both of my cars are mostly there, keeping safe distances and coming to a stop with cruise control, but definitely not completely automated yet.
Yeah I don't understand where exactly this would be usable, at least around where I live. If it's a divided highway, it would have to have stoplights. Are there places where divided highways have stop signs?
Signed, slightly-jaded person who drives the Boston<>NYC track enough to be slightly-jaded.
---
Or, Wareham -> Barnstable on Cape Cod, on any weekend morning for 6 months out of the year. Or 101 in the CA Bay Area during rush hour.
Basically, any time+place where the the thought of driving elicits an audible moan from the people then and there.
German autobahn.
And as others have mentioned it's still a traffic control device.
There's some very general info here: https://www.roadandtrack.com/news/a39481699/what-happens-if-...
I'm guessing the feasibility of this is very city-dependent. in LA, the usual traffic pattern is "drive 60-70 mph for 20 seconds, slow down to 5 mph for 60 seconds, repeat". (Granted that that's because of poor drivers, it would be better to have a consistent speed of 30 mph, but there's nothing Mercedes can do about that either way.)
Instead, Tesla fsd, at least from the youtube videos, looks like it's driving like a BMW-driver. Way way way too aggressive.
The biggest contributors to car crashes is speed and not enough distance from car in front. If self-driving cars would exaggerate the basic premises of safe driving, low speed, low acceleration, long distance, ... it would be really good for traffic overall imho.
Speed differentials kill.
That's the cope people use for habitual speeding. 10mh is not significant differential. 30 vs 60 on a highway, sure, 55 vs 65, not at all.
You should actually try it once. Go five under the speed limit and keep generous distance with the car in front of you. You'll barely notice it. Traffic will be ahead of you, you won't pass anybody. The biggest thing to get over is the ego-thing.
As you are doing this, then also pay attention to your capacity to act on any emergency stop you may have to make (dog sprinting across, car slamming their breaks, ...) and how much much more time and capacity you will have to respond.
The other thing that peoples mind immediately go to multi-lane highways. Never the other 70-80% of driving, in town, single lanes, where going slower is always manifestly better.
Also, acting in capacity to react in an emergency is more about following distance than speed. And yeah, as speed increases a driver needs to increase their follow distance. Something that I agree loads of people fail at doing and then complain about their ADAS systems always slamming on the brakes suddenly.
> where going slower is always manifestly better.
Just tell that to all the cyclists going < 20mph in 40-50MPH roads. They're way safer going that speed than those fools driving their cars near the speed limit. It's often not safe for them, partially because...speed differentials. To solve this, we shouldn't just restrict cars to only go cycling speeds, we should build infrastructure so similar speed traffic is grouped together and separate, reducing...speed differentials.
If I started driving my car 5mph in a 40mph road, I'd probably cause more accidents than if I just went along with traffic at 43mph.
Speed differentials kill.
That's again a 20-30 speed deferential, not to mention a huge difference in weight. We're talking about a 5-10 one between cars.
Also, if it's a heavy freight truck going 20 in a 40mph single lane. yeah, no issue at all with that speed deferential isn't there? Maybe the problem is here the inattentive, impatient drivers plowing through the cyclist?
> acting in capacity to react in an emergency is more about following distance than speed
The cars in front are not the only hazards.
Overall, I think you're making it too extreme. I'm not saying you should be going 20 on a highway. I'm saying going 5 under a posted speed limit is actually very reasonable, and it's what self driving cars (and human drivers) should do. It will reduce crashes. I think we disagree there.
It's not just cars that become inattentive, and I agree the ultimate fault of those accidents are with the operator not paying attention. I've have cyclists swerve in front of me pretty close seemingly unaware I was there as they cut over for a left turn at the last second without signaling. Or cyclists blast through an intersection without stopping despite me already properly starting to go through the intersection. It's not like only people in cars make mistakes. However, speed differentials still increase risks. Reducing speed differentials and encouraging everyone to go about the same speed is better than having a mixture of speeds in the same traffic flow. Mixed speeds cause friction, friction increases the likelihood of accidents.
> Also, if it's a heavy freight truck going 20 in a 40mph single lane. yeah, no issue at all with that speed deferential isn't there?
No, there's still the exact same issues with that speed differential. People bunching up, lots of changing lanes, differences in speeds, etc. Sure, sometimes equipment needs to take roads and just can't or can't safely operate near the posted speed limit. I'd still say that equipment is causing more traffic flow issues, and thus more chances for accidents, being a slow member in traffic compared to all the other cars going about the same speed as each other. But they do have a right to use the roads, and I do agree people just need to deal with the disruption and be better operators around those obstructions. In the end there's still little excuse for cutting people off and changing lanes without watching, but if they never had to change lanes...
> I'm saying going 5 under a posted speed limit is actually very reasonable,
Going 5 under can be reasonable, I agree. There are lots of instances where one can do it safely, and increase the safety of those around them. If I'm driving down a residential street posted at 30mph but there's a high likelihood of pedestrians popping out from around parked cars, I'll drive under the speed limit, often 5mph+
I definitely disagree that individuals should subject themselves to a hard rule of always driving 5mph under the posted speed limit though. It all depends on what's the reality of the situation at the moment. If everyone is already going well over the posted speed limit, going well under it isn't going to increase overall safety. If the roads would benefit from traffic going slower, the speed limit should be changed and the road should be modified to encourage lower speeds.
If the roads should be 5mph slower, we should redesign it so everyone goes slower, not just some small percentage of cars while everyone else blasts past without immediate consequence.
Could a lot of our roads be safer if we redesigned them to make people drive slower? Sure. If I cause traffic congestion and drive in a way that's outside other driver's expectations I'm not making things safer though, I'm causing problems.
And I used to drive exactly speed limit (as measured by GPS) and that maded me among the faster cars on highway. Only few cars went faster then me.
I made effort to slow down lately and can confirm that the biggest and only issue to overcome is the ego and the knee jerk "being there faster makes you better driver" kind of thinking.
I will regularly get passed by someone going 75mph or more (in a 55) in the left lane when traffic allows those speeds. It's about 50-50 whether I pass them back when we get to the stop light at the end of my exit ramp a few miles later.
On a typical commute or trip around town, driving way faster than everyone frequently doesn't get you to your destination any sooner. It does make things more dangerous and unpredictable though.
We are lucky that our ancestors were not so risk averse, because if they were we would not have cars at all, or airplanes.
That makes sense to me. I'm happy to accept the presence of full self-driving technology on the roads, for other people, once it has an accident rate comparable to or slightly better than humans.
I personally won't use one until it is so much safer than a human that the level of safety is the #1 feature though. Until then, what's the upside? I can screw around on my phone more often? I do that too much already, and it's not the kind of benefit that cancels out the potential downside of "...but you died because of an unhandled edge case in version 27.1.828 of our software that you as an attentive human would have easily handled, which was fixed in the next release" which just seems like such a banal way to go for more screen time.
I don't think my take is drastically out of the mainstream. It also seems to me the main thing separating my point of view from "ban all FSD until perfect" is a willingness to let other people make choices I don't think are good.
We can’t expect this technology to work well in the real world unless it is tested in the real world before it works well.
Tesla's system are unsafe, by default, if they don't follow safety life cycle. And they don't - I saw dick-sharing apps that had better life cycle processes.
Interventions are more nuisanced than just "Did the driver intervene". Many times I intervene and take control while using Tesla FSD not out of safety reasons, but to be nice to other drivers, or for a smoother ride. It tends to love passing other cars on the right and not letting cars into merges, for example. It also brake a little hard at traffic jams, for instance, not nearly to a point where it would be a safety issue, but when I see a traffic jam far ahead I would begin decelerating much, much earlier just for the comfort of myself and passengers.
That said, FSD is nowhere near ready, I do have a huge number of safety related interventions as well, but reducing this to a number like L3 or L4 is trying to oversimplify a problem that isn't simple.
Is it supposed to do illegal things? It shouldn't be so hard to ensure it obeys laws.
What if they disengage right before an accident in order to transfer the liability to you?
But I would expect that the disengagement is much less abrupt than what Autopilot/FSD do. From [0]:
> After consulting with the engineer in the passenger seat, I closed my eyes completely, and just eight or nine seconds later a prompt popped up asking me to confirm I was still alert. I ignored it, which soon started the 10-second countdown toward disengagement.
Which makes sense: if the point of the system is for you to be able to turn around and help your kids for a few seconds or watch a TV show on the center console, they simply can't expect that they can ding and have you regain control instantly.
[0] https://www.motorauthority.com/news/1136914_mercedes-drive-p...
I am a FSD skeptic but I might be sold on this.
[0] PDF https://static.nhtsa.gov/odi/inv/2022/INOA-EA22002-3184.PDF
"The agency’s analysis of these sixteen subject first responder and road maintenance vehicle crashes indicated that Forward Collision Warnings (FCW) activated in the majority of incidents immediately prior to impact and that subsequent Automatic Emergency Braking (AEB) intervened in approximately half of the collisions. On average in these crashes, Autopilot aborted vehicle control less than one second prior to the first impact."
Pretty sure he has the exact same attitude to the liabilities involving people getting killed by using his "Full Self Driving" beta test.
[0] https://www.therobotreport.com/un-allows-autonomous-vehicles...
> Limited to a top speed of less than 40 mph
> Operates only during the daytime and in clear weather
Wait, don’t these things together mean it isn’t useful anywhere? Limited access roadways tend to be 50+ limits, going 40 or less on them other than in the adverse conditions ruled out by the third criterion would probably get you pulled over for obstructing traffic, unless you had Slow Moving Vehicle placards.
The car almost caused an accident 3 separate times.
The first, was when it almost drove into oncoming traffic at an intersection where the road curves north on both sides.
The second, was when it failed to understand a fork on the right side of the road, swerved back and forth twice, and then almost drove straight into a road sign.
The third, in downtown, was when a brick crosswalk confused it on a left turn, causing it to literally drive up onto the corner of the intersection. Thank God there weren't pedestrians there.
When we upgraded to a Y, I turned down FSD. I don't imagine I'll be paying for it ever again.
It would've still stopped - the FSD system doesn't override the background tasks that power the AEB systems that stop the car from hitting humans or cars.
I know this was just an example scenario and your point is broader, but I'm struggling to think of another circumstance where autopilot is at fault 30 seconds post disengage, as it's effectively just a fancy adaptive cruise control.
I've never been at an intersection where straight ahead is a wrong way (wouldn't the traffic be then facing you?), but like any cruise control system, it would likely require driver attention.
Autopilot is not intended to be enable and go to sleep. It's simply fancy cruise control.
Unless you are referring to these [0] incidents from 2016, before FSD was released. That was Tesla Autopilot (which comes standard on all Tesla vehicles now).
Also, FSD uses active camera monitoring to make sure the driver is paying attention, so no you can't sleep in your while FSD is activated.
[0] https://www.wired.com/story/teslas-latest-autopilot-death-lo...
We don’t know that. An Tesla cannot claim it.
Would it have though?
https://www.youtube.com/watch?v=3mnG_Gbxf_w
At this point I can't imagine buying anything Tesla related. Or anything related to Elon Musk for that matter. He's a grifter who has managed to stand on the shoulders of giants and call himself tall.
Even when it's as bad as throwing you in the wrong lane of traffic.
This police officer was responding to an RV which had run off the road, and was speaking with the driver. The driver, a bit shook up from the experience explained that he was driving down the highway, turned on the cruise control, and got up to make himself a sandwich...
I think the Simpsons done it at least 30 years ago.
Those other cars don't claim to drive themselves.
But, he'd definitely risking other road users and pedestrians if that car keeps trying to run up sidewalks and cause other havoc on the roadway.
Now if the car failed in such a way that you couldn't disengage FSD, THAT would be a serious, catastrophic problem... but as far as I know, that's not really an issue here. (If that were an issue, obviously, it would be cause to have the whole damn car recalled.)
All of this to say, I think we can leave the guy alone for sharing his anecdote. When properly attentive, it shouldn't be particularly dangerous.
If a product being overhyped prevents you from using it after you paid for it, you're gonna have to live with no computer, no phone, no internet, no electricity, no cars, no bikes.
Every piece of technology has a variety of failure modes, some more likely than others. FSD is not likely to take aim at a pedestrian and floor it, just like your brakes aren't likely to explode, and neither of you are irresponsible for making those assumptions
No single human fully understands these AI models, and they can change daily. Yet Tesla is putting them in control of multi-ton vehicles and leaving all liability on humans with worse reaction time and little to no understanding of how it works or tends to go wrong.
I think you are making scenarios that no reasonable person would assume. There is a difference between 'getting confused at an intersection and turning wrong' and 'actively trying to kill the occupant by accelerating at max speed while turning the steering'.
FSD is a blackbox. Even Tesla appears to be unable to prevent regressions. One of the most sophisticated companies in the world can't prevent regression in a safety critical software they frequently update. Let that sink in.
I think the software is a half-baked gimmick but come on "look guys, I care about children too" variety of in-group signaling with a side of back seat parenting adds less than nothing to the discussion of the subject at hand.
And INB4 someone intentionally misinterprets me as defending the quality of Tesla's product, I'm not.
Regardless, you haven't answered the point about the quote. "Judge not lest ye be judged" does not mean we have to empty-headedly refrain from any sort of moral criticism. In context, it's about hypocrisy, reminding us to apply our standards to ourselves as stringently as we do others. I think it's only appropriate here if tsigo somehow indicated he would happily endanger his own children, which I don't see any sign of.
Deciding that you want to put your family into that steel slug for the very first test seems to me to be an entirely different level of poor decision making.
I was expecting FSD to be bad, but I sure wasn't expecting it to be that bad.
Maybe without disengaging any of the three incidents could have become problems, but for someone who knows how Autopilot works it was more comical than dangerous.
Agree raw data, or even just per 10K mile stats, from Tesla should be table stakes. Why aren't they required to report such things by law?
Tesla releasing the actual raw data would be much more helpful, but of course they are refusing to do that, most likely because it would betray how overhyped, unreliable and dangerous the software is.
Is that enough? The software could decide to accelerate and switch lanes at such a fast rate that the driver wouldn't have time to intervene. It hasn't happened yet to my knowledge. But it may happen.
As an automotive engineer: Agreed. Realistic experience reports are useful, and that includes also e.g. what drivers are willing to attempt and how they risk-rate.
I don't know what the original poster would do, but hopefully they will be more inclined to think twice next time they consider performing the behavior in question. If they do think twice and the situation is unsafe, I certainly hope they won't put their kid at more risk just to spite an internet stranger.
But my primary interest is in the many more people who might be inclined to imitate the poster's behavior when they get the chance. Having the behavior contextualized like this can only help encourage them to think about the risks.
Unfortunately, it increasing looks like the experiment has failed. But not because of us. We're pissed because Musk isn't doing his part in the deal. He's not pulling his weight. At this point, he's becoming more and more an anchor around the neck of technological progress. That didn't need to happen. It didn't need to be this way. So yeah, we're pissed off, not just because of the money we paid, but also because we feel like we were defrauded by his failure to pull his own weight in the deal.
I wouldn't be surprised to see him make his escape to the Bahamas before this decade is up.
Why not restrict your beta testing to a closed course? A race track? Some land you bought out in the middle of the desert? Have the kids make some some stop signs, plow your new roads, etc.
No one else on the road is consenting to a technology that could rapidly accelerate and course correct into their vehicle at some undetermined time.
Funding responsible self driving research seems like a great use of money to me, but testing an flawed system in the wild does not.
For those that didn't buy it, you can 'rent' it for a monthly subscription fee.
I hate to say it but starting to get this vibe too, particularly when I watch interviews from him which are 1-3 years old. They don't age well.
They're full of promises like, "only do good things for your fellow man, be useful etc" and those ethos seem to be lost now.
The fact that you associate those ideals with purchasing a consumer product made by a public company is intentional.
Enjoy beta testing FSD as an unwilling pedestrian.
(I assume this is why the Cybertruck features bulletproof glass, in case my self-firing gun (now in beta) misidentifies the tires.)
I still read this as Cyberduck every. damn. time.
>in case my self-firing gun (now in beta) misidentifies the tires.)
are you mounting that self-firing gun to a car with FSD? that would make for a great hands free experience.
That is, I fully understand that to get to Levels 3, 4, and 5, you need to pass through levels 1 and 2 in autonomous driving. But the issue is that I feel like these systems are basically at the "slight drunk teenager" stage, with you as the driver having to ensure they don't mess up too badly. Honestly, unless I can, say, read a book or do something else, these systems (I'm specifically referring to an advanced cruise control/lane keeping system) right now just require me to pay MORE attention and they stress me out more.
Fully understand we need folks to help train these systems, but, at least for me, they currently make the driving experience worse.
Speed control I'm fine with and is obvs. a mature tech that has been around for decades. Maybe it's the way I drive, but I find lane assist a liability -- especially on curves. More than once the car swerved unexpectedly one way or the other going around a bend. After the 2nd time that happened, I shut it off.
I've found adaptive cruise control to be a simple, noticeable improvement.
My five year old Honda has a very limited driver assist system (radar cruise control + lane centering), which (in my opinion) is very good at what it's trying to do. It has no pretensions of being a "self-driving" system, but it very successfully manages to reduce some of the stress of driving and make my driving better. I think the key point is it only automates the fine adjustments and provides alerts, but is very careful to never allow the driver to rely on it too much.
Until autonomous vehicles reach "read a book or fall asleep" levels, this is all I'm interested in. No thank you to any dumb "autopilot" system that I can't actually trust, but tries to control my wheel.
When I'm in stop and go traffic, I really like not having to constantly go back and fourth between the brake and gas pedal, I can just pay attention to what's happening in and around the lane and let the car keep a good follow distance.
I've gone over 100 miles at a time without having to touch the gas or brake pedal, even through stop and go traffic.
My only disdain for driving stems from sharing the road with other driver who are completely stupid and seem to lack and critical thinking skill(this shows up in especially in traffic jams and navigation within them on local roads with obstacles also present, like a parked truck loading).
A friend of mine bought a Model Y and got access to FSD about six months ago. I've spent a fair bit of time in this car in the bay area and... I'm impressed? It doesn't drive like a professional but it feels safe.
My friend says it's been improving even in just the time he's had it. So maybe it used to be a lot worse?
I'm not in the market for a new car but the next time I am, FSD is going to be a big factor. Even if it's just as good as it is right now.
I had one of the first Model S vehicles that was autopilot-capable. Early enough that autopilot itself was added later. The early versions were... intense. Not just in the "if you take your hands of the wheel it might try to kill you" intense, but also in the "even using this as adaptive cruise with lane-keeping, sometimes it will suddenly try to veer off the road and murder you" sense. Even when working "as intended" it would routinely dip into exit ramps if you were in the right lane. As a result, I didn't use it all that often, but over not a lot of time it improved pretty dramatically.
At this point my wife and I are on newer Model 3s, and our experience with autopilot (not FSD) as a frequently-used driver assist has been... uneventful? Given the _original_ autopilot experience, though, neither of us is particularly eager to try out FSD yet. Autopilot strikes a good balance for us in terms of being a useful second pair of eyes and hands while unambiguously requiring us to drive the damn car.
But that is not what we have right now. Right now every driver on the road is exposed to a potential instant tragedy that is unavoidable. I mean, what is a self driving car going to do if a large semi drops a huge log or freight right in front of you? You have one second to impact. You can either attempt to hold your line and drive over the object, potentially killing/injuring the passengers. Or you can swerve left into oncoming traffic. Or you can swerve right into a busy two way frontage road.
No matter which choice is taken, I guarantee there will be a lawsuit. Perhaps one way forward would be perhaps something similar to what medical companies did in the 1980's with the creation of the "vaccine courts". Maybe we need some kind of new "autonomous driving" court which would be staffed with experts who would then decide what kind of cases have merit. That would at least better shield the parent companies and allow them to continue innovating instead of worrying about potential litigation.
In downtown Seattle doesn't it drive into the monorail columns?
It did not work and based on my experience, I am extremely skeptical it will ever work in places like Seattle. The car could not even navigate circling Greenlake without a disconnect.
Sold them at the high of the used car market because, in part, I estimate FSD is a liability for the brand and will, eventually, hurt their resale value. That and Elon Musk. Don't need to support that and won't in the future.
(The loaners that I have had three of those four times have all had FSD.)
I am reasonably satisfied with Enhanced Autopilot on highways, though it’s unclear to me what, exactly, is ‘enhanced’ about it. And Navigate on Autopilot seems to add nothing of value.
It’s not flawless, especially on the seattle roads where knowing where the road actually is requires human level reasoning most of the time, so it drifts into parts of the road that I know from experience are for parking but the car assumes is just a widened road. Or there a lot of roads with margins that are gravel but are so subtly different from the damaged road surface it thinks that’s part of the road and tries to drift over before it realizes it’s not. These issues are nothing like the unmitigated disaster it used to be, though, and if they can keep the pace they’ve had over the last 6 months up for another 12 months, it’ll be remarkably useful.
I believe that's a "feature". I had it on a loaner too and it wants you to give it permission to cross the intersection by pressing down on the drive stalk or briefly press the accelerator pedal.
https://www.tesla.com/ownersmanual/modely/en_eu/GUID-A701F7D...
But this raises the question, what would have actually happened if you just let the car be? Would it have gotten into accidents? Or maybe it just drives in this alien unfamiliar way but is actually safer?
As long as the roads are a mixed environment of autonomous and non-autonomous vehicles, driving in an unfamiliar way is by definition unsafe because other road users can't anticipate what your car is going to do. That's not even mentioning pedestrians or cyclists.
What I realized was just that I was being scared of the computer. It wasnt about to drive into traffic, and it wasnt about to crash into anything or anything like that.
What was happening was that I was rightly being overly cautious of the beta program, and taking control as soon as there was really anything other than driving in a straight line on an empty road.
Over time, it became a dependable, usable copilot for me. I would massively miss FSD if it was gone.
It's engineering malpractice to test this product on public roads, in my view. It's beyond malpractice that the government - at city, state, and federal levels - is allowing it.
(To be clear, I am not against self-driving cars in all instances. I am talking specifically about the Tesla FSD, which has been dangerous since launch and isn't getting better.)
1) We're only going to fully "solve" self-driving with ML techniques to train deployed NNs; it can't be done purely in human-written deterministic code because the task is too complex.
2) Those NNs are only going to come up to the necessary levels of quality with a ton of very-real-world test miles. Various forms of "artificial" in-house testing and simulation can help in some ways, but without the real-world data you won't get anywhere.
3) Deploying cars into the real world (to gather the above) without some kind of safety driver doesn't seem like a great path either. There's no backup driver to take over and intervene / unstick the car, and so far driverless taxi fleet efforts have been fairly narrowly geofenced for safety, which decreases the scope of scenarios they even get data on vs the whole real-world driving experience.
4) Therefore, the best plan to acquire the data to train the networks is to use a ton of customers as safety drivers and let them test it widely on the real routes they drive. This is tricky and dangerous, but if it's not too dangerous and the outcome saves many lives over the coming years, it was worth it.
My counter-counter-point would be that there's plenty of other companies that are doing this more safely, and also that ends don't justify the means when those means involve killing pedestrians.
zoox was bought by Amazon for $1 billion, which seems a lot but it was the amount of money invested into company, so it was sold at cost to Amazon.
argo.ai just shutdown. VW and Ford spent several billions of dollars on that.
drive.ai shutdown and was acqui-hired by Apple for the car project that was reportedly just pushed to 2026
aurora is publicly traded and is on the ropes, reportedly trying to find a buyer before they run out of cash.
We'll see how long GM and Google will be willing to put ~$2 billion a year into Cruise / Waymo. I don't see them generating significant revenue any time soon.
Tesla and comma.ai have a model where they make money while making progress. Everyone else just burns unholy amounts of capital and that can last only so long.
As to crashes: the disengages part of your rebuttal is implied claim that Waymo / Cruise are perfectly safe.
Which they are not.
FSD have been deployed on 160 thousand cars. No fatalities so far. No major crashes.
Cruise has 30 cars in San Francisco and you get this:
> Driverless Cruise robotaxis stop working simultaneously, blocking San Francisco street
> Cruise robotaxis blocked traffic for hours on this San Francisco street
Another Cruise robotaxi stopped in the muni lane.
Waymo car also stopped in the middle of the road.
Neither FSD or Cruise or Waymo had fatalities.
They all had cases of bad driving.
This is not Safe-but-will-go-bankrupt vs. not-safe-but-won't-go-bankrupt.
It's: both approaches are unsafe today but one has a path to becoming safe eventually and the other doesn't, if only because of economic realities of spending $2 billion a year without line of sight for going break even.
https://www.theverge.com/2022/7/1/23191045/cruise-robotaxis-...
https://techcrunch.com/2022/06/30/cruise-robotaxis-blocked-t...
I didn't mean to suggest that. I was responding to your words here:
> Tesla and comma.ai have a model where they make money while making progress.
I'm saying that it's not OK for a car company to keep going with dangerous self driving just because it can afford to.
> FSD have been deployed on 160 thousand cars. No fatalities so far. No major crashes.
That doesn't seem to be the case[1]. Though now we're going to squabble about definitions of "major" and also how is this reporting happening.
[1] https://www.latimes.com/business/story/2022-07-14/elon-musk-...
We can't wait for perfection. The question is how much risk are we willing to absorb.
If the FSD crashes, but much less than human drivers, I'm all for it.
But how much less is enough? Very interesting question!
1000:1 sure!
10:1 strong maybe!
11:10 probably not, given the moral and legal can of worms, though it'd be nice to save ~10% of injuries...
I recall Waze had some point system to help gather positioning / speed data for side roads that it would try to have you go get with icons... and those were just fake internet points.
For example, you can imagine FSD would determine to go straight down a straight lane with no obstacles - that would be the correct behavior. Now imagine in real life the driver takes their hand off the wheel to adjust the radio or AC, and as a result the car drifts over and lightly cross the lane marker - this doesn't really matter because it's broad daylight and the driver can see there's nothing but sand and rocks for 2 miles all around them. What's the machine conclude?
Even in the anecdotes throughout the rest of the comment section, there's a lot of people that said "yeah I tried FSD for a limited period of time and it worked for me". Because we're not saying that taking FSD outside right now will kill you within 5 minutes. We're saying that even if it kills you 0.01% - that's pretty sketch.
The general principle is that all of the drivers that have been recruited to be 'teachers' to the machine are not aware that they are training a machine. As a result, they are probably doing things that are not advisable to train on. This doesn't even just apply to machines - how you drive when you are teaching your teenage child is probably different than the things that you do on a regular basis as an experienced driver. If you are not aware that you actually teaching someone something, that's a dangerous set of circumstances.
The point being made was basically that in the aggregate you can more or less generalize to something like "the tall part of the bell curve is good driving and everything on the tails should be ignored".
Since learning happens in aggregate (individual cars don't learn – they simply feed data back to the mothership), your example of a single car errantly turning the wheel to adjust the radio would fall into the "bad in different ways" bucket and it would be ignored.
I accept that as a plausible hypothesis to work off of and see how far it goes, but I would not bank on it as truth.
I'll give another example, I think a significant portion of the time, people roll through stop signs (we'll say, 25% of the time? intuitive guess). I do it myself quite often. This is because not all intersections are built the same - some intersections have no obstacles anywhere near them and you can tell that duh, there's no cars coming up at the same time as me. Other intersections are quite occluded by trees and what not.
I'm fine with humans using judgement on these, but I would not trust whatever the statistical machine ends up generalizing to. I do not think rolling a stop sign makes you a 'bad' driver (depending on the intersection). Still, if I knew I was teaching a machine how to drive, I would not want it to be rolling stop signs.
Not to discount this at all, but... yea
Even if the brains of it become perfect, I doubt the vision-only approach (or has that changed?)
They need at least somewhat decently consistent 'signal' to act appropriately... and there are some mornings I just don't drive because visibility is so poor
[1] https://www.theverge.com/2016/10/19/13341194/tesla-autopilot...
Thus comparing the disengagement rates does not make sense.
That Tesla is even allowed to do it speaks volumes to the unchecked power Elon's influence yields.
Maybe you should use specifically trained test drivers, who are acutely aware of the limitations and know how to deal with them, not random people who have been told through intentional snake oil marketing by a billionaire with a god complex who needs to feed his ego that the car can drive itself.
It's insane that governments allow these vehicles on the road.
Also, that kind of the-end-justifies-the-means reasoning has lead to a lot of catastrophic developments in history. Let's not go there again.
Anyway, the reality is that Teslas are safer than any other car on the market right now, despite the scary voodoo tech. So it seems in this case the means are also justified. If auto-pilot and FSD were causing more accidents than humans, we'd be having a different conversation about ends justifying means, I surely agree.
What data are you using to conclude that?
Safer for Tesla drivers or others near and on roads?
Customers had to opt-in to request beta-testing access, they had to pass a Safety Score system that grades their driving safety (same as car insurance apps, roughly) for a long period (in some cases many months!), etc. After going through those hoops, when they finally get the software for it, they're required to consent again. IIRC the text there includes things like: You are in control at all times, must keep hands on the wheel and eyes on the road and intervene for safety, you are liable for anything that happens, this software can do the worst possible thing at the worst possible time, etc. They also monitor for your hands on the wheel (via torque sensing) and use an in-cabin camera to monitor whether you're watching the road or looking at a cellphone, etc. These measures can be defeated with effort by unsafe idiots, but that's no different than the risks such unsafe idiots present when driving any car.
With all of that in place, they've scaled up over a couple of years to 160K customer test pilots. Accidents happen, but there's no evidence the rate of them is anything to worry about. If anything, what evidence there is seems to point in the direction of FSDb test drivers being safer than average. However, they're supposedly removing the Safety Score part of the process Very Soon (likely in the next few weeks), but the rest of the warnings and mitigations should remain.
--- meta stuff:
There's a ton of money and a ton of ego pushing huge agendas in every direction when it comes to anything Elon-related, Tesla included, especially since the Twitter saga began and he started really going off the rails more. Almost anything you read on related topics, regardless of which "side" it's on, you have to question the motive to even begin to understand the underlying objective truth. I follow Tesla news a lot, and I'd say ~90% of all random internet news articles on these subjects (positive and negative) are mostly utter bullshit clickbait when they're not outright fraud, and they're designed to influence stock prices and/or buyer behaviors more than they provide useful information. When big money and big egos are in a war over something, objective truth on the Internet is a casualty.
If you ignore all that line noise and look at the objective reality of the engineering parts though: it's pretty amazing beta software with a lot of future potential, and the testing has gone pretty smoothly in terms of safety. It could be many years before you'd let it chaffeur some elderly person on a pharmacy run as a robotaxi, but IMHO it's still a better bet than most of its competitors in the long game of fully-generalized self driving on the human-centric road networks we have today.
As for Elon himself: clearly some of his behavior and viewpoints lately are both pretty objectively terrible. At least you can see it? How many executives from companies that built things we all relied on from the past few decades have really been any better? They've mostly been better at hiding it, while Elon puts it on full display. The world is what it is.
Driving is inherently very dangerous. Traffic accidents are a leading cause of death in the US. You're not really chasing perfection to win this game. It's not all that hard to beat humans on average, because the average human is pretty terrible. It's a statistical certainty that some people will die at the hands of Autopilot even when it's in some final non-beta state, but it will probably be less people than would otherwise die to the same miles driven manually.
The hard thing for Autopilot-like systems is perceiving the world accurately. "The world" includes both the physical reality of roads+cars it senses, as well as things like traffic rules, corner cases (construction, debris, manual police control of an intersection with a dead traffic light), and gauging the intent of other/human drivers. Humans are inherently better at that stuff. The software has to get better than it is today, but it will probably never fully match the best humans at this part.
However, there are two key ways the software inherently outdoes humans:
(1) It can have more sensory data than humans. Even Tesla (currently sans Radar + Lidar, just cameras) can see in every direction at the same time, all the time, with an overlapping array of cameras. No blindspots, no temporary blindness in one direction while you crane your neck to check in another, etc.
(2) It never gets tired, distracted, or drunk. It's never checking a facebook feed, or nodding off after a sleepless night, or too engaged in a conversation with a passenger to remember to look for that cross-traffic, etc. This is a huge advantage when it comes to accident rates.
Bold claim. We would need the data to be sure. Judging by reports of Tesla owners in this thread I'd guess FSD and Autopilot are probably causing more harm than it's preventing.
> It never gets tired, distracted, or drunk.
Which would be a great benefit if FSD didn't drive like a drunk teenager.
Look for any tool humans rely on there must be predictability. And until we have enough public data no conclusions can be drawn. That's why it's in Tesla's interest to continue releasing less and less data except for the data that makes them look good.
However, this is copy-pasted from Tesla's website:
"Tesla cars come standard with advanced hardware capable of providing Autopilot features, and full self-driving capabilities—through software updates designed to improve functionality over time."
Who's to blame for this misunderstanding?
It is very bad at handling ambiguity when the road has both white and yellow navigation lines. There's a number of scenarios, some of which involve criss-crossing lines, some where they run mostly in parallel.
The car sometimes wants to follow the white lines when the yellow are actually overrides. When the yellow lines are wrong (and should rightfully have been scraped off), this always leads to disengagement. For stretches where the yellow lines are correct, and the older white lines are off by half a meter so that there's effectively not enough space between the white lines and the concrete side, autosteer will sometimes try to switch to the white lines, which would lead to the car smashing into the concrete side.
And sure, autosteer != FSD. But it's a much simpler problem that the Tesla still basically fails at.
Sometimes, in fog, cruise-control will abrubtly break. Very nerve-wrecking.
Relying fully on vision has its downside. Listening to Kaparthy on Lex Fridman's podcast, it sounds like this was cost-cutting. I can't believe that "nope, that big white thing isn't a wall" is not objectively better navigation.
But rolling your own ADAS and FSD might result in having spent a lot for a result that is inferior to what ADAS suppliers can sell to other car OEMs, with no substantial benefit to the FSD effort, which is much riskier, with the added, and perhaps gratuitous risk of trying to do FSD with only camera sensors. The risks in that are not just it is a risky approach to real time sensor data, but that the data all your cars are collecting is worth a lot less for not having a LIDAR point cloud providing 3D data.
It’d be cool if there was a yearly competition to rank the basic lane keeping/speed matching/turning. Eventually when it’s actually full-self-driving they can rank FSD.
It got a lot better over the 2018 thing, that tried to crash itself on several roadworks, I actually almost scratch the car in first 100km I had it because of it.
My latest BMW works perfectly fine in German roadworks with bright yellow lanes, Also all the stelantis cars I've driven (all 2020+ models) worked fine in that conditions.
Camera system has problems when there's fog, or direct sun in precise angles. Radar has problems where there's lot of rain, and rainsplash from trucks, or snowy conditions, and the front radar gets frozen.
But it's awesome as assistive system.
Do you do 100% of training on data from the same region? Are there low-level skills that all translate and don't bring the risk of training contradictory expectations? Can you just annotate training data with the rule set/location that applies?
You mean the left?
Great car or great charging network. Pick one.
If you pick charging network get a Tesla. Otherwise get something else.
I really want to sell my Model S (for like an Audi for example), but no one can compete with its battery:
https://docs.google.com/spreadsheets/d/1k1DOw-NwvW8E8tQeXlac...
[1] https://www.tesla.com/support/non-tesla-supercharging
[2] https://cars.usnews.com/cars-trucks/features/superchargers-o...
Tesla chargers are not open, in the slightest. Them publishing specs online on the physical plug design doesn't make it open to everyone. There's still lots of Tesla IP that covers those specs and designs. Tesla won't license it to other automakers unless other automakers essentially never enforce any of their IP.
Autosteer on our Tesla Model 3 is actually pretty great. I use it all the time because it makes driving a lot less tiring, especially over long distances. I don't know where you get your information.
As far as competition, our other car Kia Forte 2022 now comes with Autosteer like feature as well. With that, unless, the lines are pristinely painted, you are literally taking your life into your own hands.
I find it to be a fun, well performing and comfortable car. I am shopping for a second car currently and have yet to see anything close. There are no other electric cars on the market that tick the price, performance and range boxes, before I even start looking for the creature comforts I'm used to.
Cruise-control: In heavy traffic, I don't use it. I like to contribute towards reducing congestion, and the cruise-control accellerates too much at once, causing lumps. But for non-heavy traffic it's pretty great. The "braking for things that don't even exist" is a separate concern, IMO.
Ultrasonic, radar: Yeah, I'm disappointed here.
Rain-sensing wipers: It actually has that, and they do sort of work. What's annoying is, apparently in Model Y, you can't disable automatic wipers if you're on autosteer. So when you're on a very regular stretch of road, and the rain-sensing wipers think it's raining and it's not, you can either choose between autosteer + wiping a dry windshield at the lowest frequency, or driving manually.
Automatic high beams: They work pretty good in my experience.
Carplay: I get that Apple wants to be my car's operating system. It may seem great, but I'd prefer if my devices got along without total platform dominance from one party.
Because what happened really surprised me: Entering a construction zone the assistant took the wheel away from me(!!!). I was steering manually, normally, but the assistant insisted I follow the previous line in the road, that lead right into the construction zone. Which was behind a concrete divider.
I actually had to fight the "assistant", with not insignificant use of force on the wheel, to not crash sideways into that divider wall.
They've publicly admitted that the entire autosteer codebase hasn't been touched in over 4 years, since they fully pivoted to getting FSD in a usable level 2 state.
Finally comes deployment: no driver, paying passengers, remote monitoring.[1] Three companies are far enough along for driverless car deployment in California: Waymo, Cruise, and Nuro. Waymo is going about 13,000 miles between disconnects now. The remote monitoring center can reset.
Waymo is still being very cautious. Only in Phoenix, AZ is the service really deployed. There's a service in San Francisco, but you have to sign up as a "trusted tester" and it's mostly Google employees. When it goes live in SF, then this is real. Waymo has a new vehicle they're showing. It's an electric van with no steering wheel. Not clear how far away deployment is.
Tesla isn't even trying to qualify for autonomy in California any more. They've given up. They used to whine about being "over-regulated". What they hated was having to report all disconnects and accidents to DMV, which publishes them.
[1] https://www.dmv.ca.gov/portal/vehicle-industry-services/auto...
[2] https://www.theverge.com/2022/11/21/23471183/waymo-zeekr-gee...
Sunlight as a disinfectant indeed.
(╯°□°)╯︵ ┻━┻
Are you telling me there are no American companies they could have partnered with? Why would you willingly give US dollars to the CCP for a free market R&D effort?
I've been using autopilot on the freeway for years and that's been mostly fine—but I'm never going to use FSD on streets again.
This is unbelievably terrible though. I really regret purchasing it.
The one time I really beta tested a for-profit product, not only did I get it for free, I actually got a rebate on the final product (it was pycharm, and jetbrains gave me a free license for a year, which they got back many times over as I renewed yearly ever since).
Though I guess the early accesses I got for kickstarters were kinda like paying for a beta in a way.
Which in turn makes everyone else around your car an early adopter?
Which is good because a student in my group nearly killed us trying to merge into a semi.
Even when they actually admit that they have failed at it [0]. I am not sure if they are aware of the doublespeak in this admission.
[0] Failure to realize a long-term aspirational goal is not fraud.
Normally, each person puts one new driver on the road per-lifetime.
When you beta test a baby driving robot, you’re now at two new drivers per life! And the Tesla doesn’t seem to be learning faster than a human!
I suppose everyone should just assume a Tesla is about to do something silly and drive defensively ...
If we assume I just got repeatedly unlucky and this was a 1:1000 situation would that be acceptable to you?
As a specific example, when the car suddenly slams on the brakes, especially at slow speeds, it doesn't look like a big deal on these GoPro cameras - but it feels like a much bigger deal when you're in the car and you feel the g-force of your body reacting to the sudden unexpected deceleration. Some of the more transparent and honest reviewers like "Chuck Cook" even add an overlay showing these forces in real time on their videos - but seeing a number briefly spike is a very different experience than feeling it live.
I would say FSD is getting closer to "safe" (not there yet though imho) but it's still very far away from "comfortable".
There are some very obvious things it still struggles with like construction and roads without clearly defined lines. But I just don't use it on those routes.
I think Tesla made a mistake trying to take a "boil the ocean" approach that will work literally anywhere instead of focusing on key areas first. It ensures you need to 100% nail it before it can be considered "released" as opposed to releasing with a specific set of criteria to meet.
1. I have many many multi mile 0 intervention drives
2. it is getting a noticeably better every release
Their video model will be very good in a the future if the current release pattern keeps up. Not sure what it will take to be certified by the USG as level 5, but it seems like it will be at the top of this chart in the future.
That's the same reasoning as thinking a faster CPU or more RAM would solve problem.
More data could even lead to worse models.
It appears they’re abandoning the video-only approach:
https://techcrunch.com/2022/12/07/tesla-appears-to-be-turnin...
We want this data!
An equivalent improvement in driving safety might also be had by eliminating drivers under 30 and since those aren't Tesla drivers (the average Tesla owner is a 54 year old white man making over $140,000 with no children) this self-driving malarky would have zero safety impact.
With two exceptions, all of my disengagements have been "quality of life" disengagements where I disengaged for reasons other than safety:
* to take a different route or get in the correct lane * to be polite / courteous to a pedestrian who wasn't yet in the crosswalk * to move more quickly through a construction zone or other traffic irregularity.
In none of those cases did I believe that the car was going to cause any risk of an accident, at best there was risk of pissing someone off (which, to be fair, in a city like Miami with far more firearms than judgement could be fatal, but not due to car at all). So yes, these are disengagements, but they weren't dangerous.
There is one case where I do get dangerous disengagements. At the intersection pictured below, when I'm in the position of the white van, sometimes tesla will mistake a green on the far traffic light (which is for the other road, intersecting at a sharp angle) for my road and proceed if it is green even though it is red. Happened twice, haven't noticed on the last version yet. Intersection here: https://www.google.com/maps/@25.750348,-80.2061284,3a,75y,26...
* Critical: Safety Issue (Avoid accident, taking red light, unsafe action)
* Non-Critical: Non-Safety Issue (Wrong lane, driver courtesy, merge issue)
Looking at the original data sources [1], FSD seems to be improving over time at the metric that matters most:
* City miles per CDE have gone from ~50 to ~120.
* % of drives over 5 miles with no CDE have gone from 72% to 93%.
Personally I have it, I don't find it enjoyable to use -- but I also don't feel unsafe when using it. Highway autopilot on the other hand I find immensely reliable and valuable.
[1] https://www.teslafsdtracker.com/ https://twitter.com/eliasmrtnz1
Fred was an early Tesla fanboy and investor; he used to gush over Tesla and Elon Musk in his writing at Electrek. It seemed like most of the time when Tesla was mentioned in an article, he also mentioned TSLA, the stock ticker...but he didn't seem to do that as regularly for say, Ford or GM.
It was rumored that Fred had a direct line to Elon and there were also occasionally public Twitter interactions between the two, but Fred gradually become more critical of Elon, especially over FSD stuff. After one critical article, Elon blocked Fred on Twitter: https://twitter.com/fredericlambert/status/14176561698297487...
Fred certainly hasn't forgotten about it: https://twitter.com/FredericLambert/status/15186308301381795...
So, Tesla fanboy/blogger comes up with some valid criticism, Elon cuts him off, which in turn probably makes him more critical of Tesla/Elon.
I tend to drive on 95 north, in Brickell, Downtown, Little Havana, on the 395 and 112, Miami Beach, north Miami (up to 100th), and to the airport. If it can handle these places it seems like it's something to consider.
For the most part it works fine in Brickell and downtown. I put a link to a street view of one tricky intersection above. It also struggles here when you're trying to go south on Miami Ave:https://www.google.com/maps/@25.7626232,-80.1929899,3a,75y,2... basically where the black car is going south it needs to follow the street in a 90deg curve to the right then immediate left at the stop sign. Sometimes it tries to just go straight there, i.e. to where the truck is.
Also, i do often need to "intervene" by pressing the accelerator since the car is a bit to passive for the typical lunatic miami drivers.
I routinely have to take control. It definitely accelerates to hard on some sleepy side streets. It's extremely conservative about speed and lane changes. It's terrible at anticipating slowing where humans obviously would, like brake lights stacking back toward you on a curving highway: the driver can see the braking problem coming from a thousand yards ahead, maybe miles. Summon is slightly better than a parlor trick. FSD is approximately useless in parking lots.
But....
It's not completely unusable. There are definitely segments of a drive I can mentally relax (back to my normal driving level of attentiveness).
Having gotten two kids through their driving tests recently, I'm struck how much FSD 10.69 resembles a 16 year old with their early driver's permit. They're terrible. But they are learning really fast.
Similarly, Tesla is now getting millions, if not billions, of disengagements coming back for retraining. The next release is going to be better.
I'm not sure if this was intentionally funny but it made me laugh.
"Disengagement" is going to happen for different reasons for all of these different cases. There's the axis that runs from merely inconveniencing others (e.g. "I disengaged because I was holding up traffic by being embarrassingly slow through a turn at an intersection") to the ones where a serious accident was averted (e.g. "I hit the brakes moments before the car caused a head-on collision). There's the scenario differences that feed into thresholds for deciding to disengage: professional safety drivers on planned tasks that have been given a sort of "rules of disengagement", vs everyday normal humans getting groceries and intervening whenever they feel like it for whatever reason. There's the fact that many competitors are operating in tight geofenced, HD-mapped areas, while FSD is operating real-world all over the country.
I agree that Tesla's lack of transparency is troubling here, but this article seems to be trying to pressure them to increase it by taking a dishonest tack and stack all the unknowns against them in the worst possible way.
They have published basic accident statistics (in the past), and those have generally shown their automation to be a net win. In their own published stats from Q4 21 (latest available on https://www.tesla.com/VehicleSafetyReport ), the comparison they make is this:
* NHSTA data says the whole US averaged one accident per 484K miles.
* All Tesla cars on the road averaged one accident per 1.59M miles (so, ~3.5x better? This could also be caused by the profile of Tesla drivers rather than any real car safety difference - income, where they live, etc)
* Tesla cars with some form of Autopilot engaged averaged one accident per 4.31M miles driven (~2.7x better than non-Autopilot Tesla data, and overall ~10x better than the national average).
Is that a typical delay (e.g. were Q1, Q2, Q3 21 reports not available in December of 2021), or have they stopped or suspending publishing the reports?
I'm just not familiar enough with the site to know what is normal.
> Tesla cars with some form of Autopilot engaged averaged one accident per 4.31M miles driven (~2.7x better than non-Autopilot Tesla data, and overall ~10x better than the national average).
Do you know if "accident" encompasses all accidents, or is there a threshold level for severity? I'd be curious if the same patterns hold for "serious accidents" defined as someone (in any vehicle in the accident) being seriously injured or killed.
> Comparing Telsa's disengagements for its FSD beta, which can drive anywhere, against autopilots that work only on (certain) highways is not very comparable. At least compare highway to highway disengagements.
I own a Tesla, but do not have FSD (Not interested in it at all beyond being a party trick). After 3 years and 20,000 miles, I can say that I've never had to disengage AP because of imminent danger, though I HAVE had many times where the lane is splitting and AP can't decide which way to go, so I had to take control. Likewise, when a lane is merging, it would sometimes jerk left and right a bit and I'd take over. Both cases were more for comfort than safety.
Total, I'd probably average 5 miles per disengagement, but that's just so I can change lanes since the base AP does not include lane changing.
If I can't trust AP with simple road rules, then why bother upgrading to FSD?
Of course it's not 1:1. AP has other features, such as a much nicer UI that better reports what it thinks it sees, and I think has sexy lane changing features and such.
But it makes me feel that FSD was more business necessary to keep a competitive advantage against cheap cars. Soon many cheap cars will have AP-like features at comparable fidelities.
I'm hoping once the two stacks are merged, that will improve.
The other problem I have on high ways is I can't really use it in stop and go traffic. If I do, it will frequently accelerate to fast from being stopped then have to break hard when traffic reaches a stop again. Too jerky for my comfort.
Overall, when I can use it, it makes driving less stressful. I drive 600 mile round trips a few times per year and am able to have it engaged vast majority of the time on highways.
FSD Beta on city streets requires too frequent overrides for it to be anything more than a party trick at the moment. Even in very rural areas it will have problems of either not changing lanes when it should or changing lanes when it shouldn't.
Auto parking worked great for me initially but something changed and it is never able to detect parking spots for me. The few times it does it just flickers on the screen then disappears, resulting in me not being able to use the feature. I'm not sure if this is because of a problem with sensors or if it's related to me being in the FSD Beta. I've been in the FSD Beta for over a year now.
110% this.
In addition to this, I often find that it is a bit timid on the acceleration once traffic starts really moving. Like, it'll jerk a bit forward from a stop, but then veeerrryyy slllooowwwlllyy speed up, even once traffic ahead is over 100 feet away.
This suggests a trivia question, which I do not know the answer to: what lane in the US goes the farthest without a split or dead-end? Similar question for other countries.
I'd count most freeway exit ramps as splits from the adjacent lane, but would not count intersections with cross streets as splits unless the two roads intersect at a small angle. It is the actual lane that matters, not its name, so if a lane has a name change when it crosses some political boundary it still counts as the same line.
My first guess would be it is on some long freeway like I-10. My second guess would be it is on some rural road through the middle of nowhere.
I-80 starts in San Francisco and ends in New York City, but I'm certain there are splits that would make you change to some other highway if you stayed in the left lane.
Likewise, I-90 starts in Seattle and ends in Boston and travels through Montana.
My guess would be I-90, just because of Montana.
The car just does not behave like a regular driver in any capacity. It's a neat trick to show when there is nobody on the road, but besides that, I have lost all faith in FSD ever coming out in any meaningful way. I only paid $2k for FSD ($7k total w/ Enhanced Autopilot), but even that is too much for what FSD actually is.
That said, it still does dumb things like:
* Getting in the left-most turn lane when I will make a right turn immediately after a left. This usually results in a disengagement or me having to change lanes immediately after turning.
* Changing lanes, but the process sometimes makes it so that I change lanes in an intersection. This is illegal.
* Not stopping at the crosswalk on the right-most lane on a red. I expect the car to stop firmly at the intersection and then slowly creep out.
"By the middle of next year [2020] we'll have over a million Tesla cars on the road with FSD hardware feature complete at a reliably level that we would consider that no one needs to pay attention. Meaning you could go to sleep. From our standpoint, if you fast-forward a year, maybe a year and 3 months, but next year for sure we will have over a million Robotaxis on the road. The fleet wakes up with an over-the-air update. That's all it takes."
If I need to do something on a drive I get my wife to drive.
I honestly do not understand the obsession with self driving or driver assist.
I drove a Ford with active lane assist and having the car physically move the steering wheel is the worst.
But apparently Im one of those weirdos who actually enjoys driving.
But I tend to believe Gill Pratt, at Toyota, who says he doesn't know how to make level 5 AVs, more than I believe Elon.
https://www.theguardian.com/world/2020/mar/16/how-helsinki-a...
AVs can also solve other problems like efficiency and surge-ability in public transportation. Not just a toy for the rich.
I also enjoy driving, but I love the adaptive cruise control in my Toyota, because it makes the most frustrating kind of driving (lots of traffic on the highway) super easy to deal with. It makes me much less hesitant to make the trip from Austin to San Antonio, because rather than having to be hyper-vigilant for two hours due to constantly shifting stop-and-go traffic, I can let the car control the speed and instead focus more on watching out for crazy drivers.
I am not sure whether I feel the same way about full self-driving. At that point, in my mind, you'd be better off taking a train, since you're very likely to become super distracted/tired with nothing to focus on. That said, if it makes the roads safer in aggregate, I'm not against it.
Given the kind of information we see in this article and other factors, though, I'll never buy a Tesla. I'll wait until it hits the major car manufacturers.
That is a very subjective description. One fatality per 100,000,000 miles actually seems pretty good, given how complex driving can be. And even then, that number includes the shitty drivers that account for most wrecks. The median driver is quite good, objectively.
Relative to trains, buses, and airplanes, cars are super dangerous.
Relative to motorcycles, they're safe.
I feel bad for Tesla owners who are stuck with shitty cameras. Lex Friedman should have given a way bigger tongue lashing to kaparthy for being so stupid.
Self driving would be a complete game changer for the disabled.
Tesla may not ever get FSD working with its current approach, but if a competitor can do it (even geofenced like how Waymo and Cruise operate today!) and the economies of scale could kick in to make it more affordable than Ubering everywhere, then it would be a huge benefit to those that are even unable to use public transportation.
If I can be allowed to grossly generalze here: would the elderly and the disabled be in a position to afford a vehicle as expensive as this? Would they /need/ a permanent vehicle to take them to wherever they need to be, if they already have everything they need right there (hospice, retirement complex etc)? Remember: I'm grossly generalizing here, but I think my point stands.
The way I see it: I've been making use of FSD vehicles to take me to work and back for almost a decade now. Sometimes even the vehicle arrives right to my front door and takes me directly to my destination. The fact that it's wetware instead of software has no material impact on my experience or expectations. Thus I'll always go for the cheapest option.
In my view FSD is trying to solve a problem that only exists in the mind of the isolated techie who forgets about the real world sometimes.
No offense but do you own a car? Or do you only ever take public transport and taxis / ride-shares all the time?
Being able to do things on your own, when you want to, on your own schedule, especially in places that are not cities and have much worse infrastructure on both public transit and ride sharing, is immensely freeing.
FSD is far less impactful on city streets as I enjoy city driving, but it gets mind numbing on long highway drives.
Ha, AP puts me in situations (especially at merge points) I'd never put myself in. It certainly doesn't make the drive less stressful. Quite the opposite. And my wife won't even let me turn on AP now when she and the kids are in the car (this would be most road trips) because it's startling when the car brakes suddenly for an overpass.
When I think about the people I know who died under age forty, I think it's majority car accidents. At least a plurality.
This holds roughly true in the statistics: car accidents have been at least in the running for leading cause of death for Americans under 30 for years. We need to address this issue, and self-driving cars are rapidly approaching viability as a solution.
Even if we don't eliminate human drivers, having viable alternatives means we can be strict about who drives. Right now, for so, so many people, driving is a necessity, and thus we're unwilling to limit access to people who are actually good and safe at driving. If you get in an accident, your insurance company will your rates, but you keep your license. If you get in another, we'll...they raise them some more more. It's really quite difficult to actually lose your license in the US. My friend had to get a breathalyzer after multiple DUIs, but he got to keep his license through all of it.
If driving were seen as an optional privilege, we could limit it to people who do it safely, and perhaps those people would only do it while sober and alert given the option to use self driving when they aren't.
What's the world coming to when some schmuck on public transit can do things like play games or even get work done where a driver can not do that? What's worse is that smart phones make public transit a lot more usable and attractive in general. Self driving cars keep the automotive transportation mode more competitive.
What a lack of imagination.
I had already known that Tesla drivers using AP was a major cause of slow, too-far-following-distance driving (which is why I take every opportunity to cross in front of a Tesla doing 60ish), but this was a whole 'nother level of literally "fuck you, pay attention until you get home, asshole".
/Rant.
P.S. Tesla drivers are absolutely the new Prius drivers.
Edit: and the fact that the behavior of "if there's a car following don't slow abruptly" isn't a 0-th level if loop in the FSD algorithm is completely inexcusable. Shit needs to be regulated now.
In my experience, it looks like Tesla’s recent concentration on surface roads has come with some neglect of highway driving.
The fact that it's only in the last year or so that Stability/OpenAI and efficient hardware like before mentioned processors that can produce relatively fast and accurate results in the NN sphere makes me wonder if it's just "too early" to have self driving cars?
ChatGPT still requires a supercluster to run and as a layman one would think video analysis with near zero latency would be even more demanding.
Also how does it compare to to apples newer Arm based architecture and Neural Engine which as far as i know is a good example of in-house designing (not production off course).
(Running ChatGPT does not require a supercluster, training the model does)
But as a lawyer once said, "I convict with anecdotes."
In any case the reports are interesting because with them you can start to understand the shape of the autopilot's capabilities and limitations. IMO the individual reports (like the ones here) are nice and detailed; the kind of summary data the article wants is pointless, because it doesn't include any real detail in it. Your autopilot worked on a 400 mile stretch of empty Kansas highway, big deal.
I feel like a good way to play the market with all this is to look at mainstream media news service articles about TSLA before the Twitter purchase and after the Twitter purchase and train them to spot the sentiment difference in the before and after articles. It would serve as an early warning sign that the company has been switched from being boosted or trashed.
One funny thing is before the Twitter purchase, alt-right outlets like ZeroHedge were constantly mocking Tesla and predicting its imminent demise during its gigantic run up while mainstream outlets were throughly praising it. Now, it's the other way around.
As you point out, the stark flip-flop of roles between left and right-leaning coverage of Tesla is almost hilarious when observed from any "neutral" point of view. Even in this HN comment thread, the emotional injection of political points throughout is both funny and disheartening.
The Twitter acquisition (starting from his initial offer) brought expectations down to reality for everyone.
Please note that his behaviour as a multiplier (and the changing nature of that multiplier) of investor value has zero bearing on whether or not the cars he sells are any good. Just like the quality and quantity of those cars had zero bearing on TSLA's stock price.
(1) Here's what I think is the case for Waymo's operations: - Phenix AZ only (maybe one more city?) -- Amazing sunny weather -- Clear line markings -- Can overfit to said city with HD maps, handcoding road quirks, etc - Waymo employees only -- Not to sound to tinfoil-hat, can we really trust this data - Even within Phenix, some filtering will happen as to which route are possible
1. When it works it’s just great. From maintaining lanes, speed limit, distance to other cars, auto park and summon are delightful.
2. Auto park in home garage is a bit too slow to use. But parallel parking is cool.
3. Even without a radar it worked on most roads and different lighting conditions.
4. Random speeding/ slowing is definitely present. Not like jamming into the vehicle in front of you but sudden acceleration can surprise. More dangerous was random slowing down at traffic signal, even when green and at traffic circles.
5. Overall it’s definitely good enough to show promise of self driving. They are not there yet but I would place them as one of the most advanced considering the range of functionality.
6. Still the $15k is simply way too much. You are better off with occasional $200 per month like when going on a long trip.
I still haven’t been able to digest how Tesla can have beta testers pay them vast amounts of dollars for improving their product, without the ability to even opt out of omnipresent and opaque data collection… For a safety critical product that the company rejects liability for.
The old trope “you’re the product” is not enough to do justice to this bizarre situation. It’s like the peak Apple fanboyism a la 2010, but on meth.
All that's said from someone who otherwise loves their Tesla. Has saved many thousands on gas. Hasn't had any reliability or "panel gap" issues. And would likely buy a Tesla again.
Where the moat? Traditional car makers are catching up on electric and already have superior build quality. And if anything, Tesla is behind on FSD.
If superchargers are Tesla's main advantage, they are not only in competition with auto makers but with all electricity providers in an increasingly competitive market.
Tesla will be fine.
Maybe the gov't will squash FSD and Tesla will have an epiphany and decide to refocus their efforts on the basics where they're quickly falling behind. Hell, maybe they'll start including the better technology the competition relies on, and try to stick to just the things they're good at. Which is ... I don't know. Image seekers?
I can say that after experiencing the software of the BMW i4 and the Audi e-tron — holy cow, Supercharger is certainly not the only advantage of a Tesla. Audi's is laughably bad. BMW's is better than Audi's. Both are much worse than Tesla.
The obvious retort there is: but Tesla's competitors will get better. And sure, they will. You're right. But Tesla likewise has their own R&D investments and unreleased products.
I think sometimes when we talk about competitive landscape, we tend to say "Company B will catch up to Company A" while in our minds falsely imagining Company A as being static. Both Company A and B are moving, but in Tesla's case, they had a ten year jump on tightly integrated hardware + software. I think it's likely that continues to yield dividend.
It's literally just a good battery and good charging. That's all Elon has, and that advantage will die in 5 years.
Given the current rate of improvement it's hard to see how the remaining 4 orders of magnitude will be solved in a reasonable time frame, without a radical reimagining of their entire tech stack.
I bet this is extremely common.
To be fair, we've never seen a test of Waymo or Cruise on the same road, as they are only available on certain specific roads that have been well tested. I'm not making a judgment here as to whether that is good or bad.
The fact that some of his first FSD trips showed major issues is very telling.
And, tbh, those Chuck Cook videos show FSD doing some really dumb things fairly often.
Is that the reason? Not a clue! Am I optimistic about Tesla's FSD? Not even slightly! But the "data" here is extremely shaky and there are many possible explanations beyond "getting worse".
But is this why I am seeing this article on Hacker News now? Or am I seeing it now because Musk pissed off a bunch of people who disagree with his political stances?
I trust the lane-keep feature on the old 2017 Corolla I rented more than FSD.
Musk is no Steve Jobs, despite clearly seeing himself that way. Full Self Driving will get nowhere under Musk.
> his main asset is being a tyrant and taking the surplus value of the 80-100 hour weeks his peons do for their lord.
Cool, I agree, can we start holding everyone accountable that does this? Can we maybe start with the ones that use actual slave labor? Where are the dozen articles a day about those people?
> Full Self Driving will get nowhere under Musk.
It's gotten farther than literally any other organization on the planet.
Hasn't yet. And IF that happens, and if it turns out to be SpaceX that enables it, thank Gwynn. Musk is many things, but this idea that he's some kind of genius engineer is extremely insulting to all the actual rocket engineers.
They are mostly for gullible people who want to pay for fairy dust. "There's a sucker born every minute."
Let it be ready, then I will see reviews and only then will I pay a bit to get it.
If they are developing it and it is getting worse, this simply means there is a lot of technical debt and more to surface. The guy leading it has left as soon his stock vested, why not stick around and write history?
Probably when waymo departed from them and they thought they will fork the code and then improve it is where it all went wrong.
This also means that new devs and engineers cant make sense of the code and product, we have all inherited some of these code bases, havent we.
Seen many such people who deploy much less than a mvp and then face the consequences eventually.
All the mvp out there who survive the beginning at least have a solid software architecture.
Guess what: I like regular autopilot better!
We should let just anyone fly planes. And when all those planes crash, we should point out that it's not fair to show that planes with properly trained pilots don't crash. Not all airmiles are equal! Those non-crashy airlines only use pilots with very specific training. Not the same thing at all!
Source? Do you work at one of those companies?
"Next year for sure, we will have over a million robotaxis on the road," said Musk on October 21, 2019. "The fleet wakes up with an over-the-air update. That's all it takes."
https://twitter.com/PPathole/status/1249209968877862919
Elon Musk @elonmusk · Apr 12, 2020 Replying to @PPathole and @Tesla Functionality still looking good for this year. Regulatory approval is the big unknown.
The biggest problem with FSD is bounding box detection as from previous videos it routinely fails to identify objects in the road correctly.
A problem that LiDAR (which Cruise uses) is especially good at.
You only have to look at Cruise versus FSD videos to see how the latter struggles with reliable object detection.
Also this is about Cruise the robo-taxi company and not the GM SuperCruise technology which isn't what I would consider a peer to FSD.
No thanks.