Wow incredible work by the team that decides who can access the FSD they paid for, very impressive. Now let's see if the engineering team can match that with a product that works.
It's a commercial name for a Level 3 autonomous function on Tesla cars ("the car’s systems perform the entire dynamic driving task (DDT) within the area that it is designed to do so. The human driver is only expected to be responsible for the DDT-fallback when the car essentially “asks” the driver to take over if something goes wrong or the car is about to leave the zone where it is able to operate." [1]), considering that the "area" is not strictly defined, and could be any road with sufficient visual cues for the car to assume where a viable lane could be.
In terms of sensors, it relies exclusively on optical cameras, as opposed to other autonomy projects that use LiDAR and/or radar and/or ultrasounds as well as cameras.
People have been paying thousands of dollars to subscribe to FSD for many years now, even when the only autonomous interventions in the car were strict Level 2 autonomy. Contributing factors for this seemingly irrational investment could be that the Level 2 features worked great (except for phantom braking and other arguably dangerous mistakes), and the backhanded marketing that would place FSD as a Level 5 autonomy feature.
As I understand it this is Tesla's hypothetical Level 4 automation. Except Tesla doesn't really offer that, so this "beta" is really Level 2 but sold like it's a bigger deal.
SAE automation level 2 means the machine can do some of the boring parts of driving, but you as driver are actually always responsible, you are required to overrule the machine whenever it does something which was a bad idea and that's somehow not Tesla's fault. If you somehow fall asleep while a level 2 car is "driving" you're possibly never going to wake up, and if you do chances are any collision or other damage is your fault.
At level 4, the machine will either do its job or safely give up. So it's OK to be asleep, although it's possible when you wake up you're like "God damn it car, I wanted to be in Boston by now, and you've just sat here moaning about a little bit of snow just outside Plattsburgh all night"...
(At level 5 the machine should drive in any reasonable conditions, this probably gets very elastic about what constitutes a "reasonable" condition for driving - is this car level 5 despite the fact it won't drive in a blizzard? Maybe not driving in a blizzard isn't cowardice but common sense? A dirt track seems fine in a 4WD with AI, but how about this boulder-strewn mountain path?)
Does this beta has any roadmap to become "stable"? The issue with FSD is the corner cases and I doubt we will see real FSD anytime soon. I would say the beta FSD is the equivalent of an intoxicated driver.
My personal guess: if Gmail stayed in beta for 5 years after it was opened to the general public, and it was "just" an email service where the only unknowns are how people and servers will abuse the SMTP protocol... then Tesla's FSD will probably stay in beta forever :-)
Of course it will, just not as soon as it was advertised. The problem is not that FSD won't be successful, but that many people were led to believe that it will be done in 1-2 years if they just buy the software package.
If you're going to claim that first video was 'doctored', how about the second, the follow-up in a school zone, on a crosswalk, with the child dummy wearing a reflective safety vest, showing FSD enabled, and the accelerator not being depressed, with everything in view at all times?
>1) not fsd, just a test of automated emergency braking
So you're saying we should trust full self-driving on a car that can't brake in emergencies reliably? And that this is ok?
>2) lidar guy. You'll see a ton of responses to his antics on HN and YouTube.
From fans of Tesla & Musk. Just as untrustworthy. Except one side here has provided evidence of a problem, and the other side has responded by making excuses & attacking the guy.
Yeah, we have those things called safety regulations that protect people and don't allow anyone to just deploy whatever software they call "full self driving" that doesn't really full self drive the car.
Except regulations have nothing to do with this as Tesla doesn't follow those. Tesla update their cars with software changes in areas that by law requires the cars to be re-certified. For example if I go out now and do something that add 50 hp to my car or change the way its braking system functions (even if it makes it better) it is instantly illegal to drive the car until it has been approved (like an MOT test in the UK). Tesla don't give a donkeys arse about those rules and no government so far have dared do anything about it (Sweden almost did, then didn't). All updated Teslas are illegal in the EU as far as I can tell and that isn't even considering the whole GDPR can of worms that is their cameras.
As someone who has been following Tesla FSD progress through the years through some YouTubers[1], the progress has been very very impressive.
2 years ago, FSD was pretty awful. Back then, my opinion was that Tesla FSD was doomed: Self driving without LIDAR was not going to be feasible for at least a decade. Now, I've changed my mind. Current version of course has a long way to go, but the current state and the rate of progress makes me believe now that, in 2 or 3 years, with a new hardware update, Tesla FSD will be very successful.
I recommend checking out some videos of the latest version (10.69.3.1) to get an idea of how good (or bad) it currently is.
I don't understand this attitude. Shouldn't the requirement be that it's better than the alternative, i.e. human drivers? Human drivers certainly aren't impeccable.
It probably depends on how well current hardware performs. If it's good enough they might not (however if you have a newer car with newer HWU FSD will perform better of course).
Camera placement (A-pillar camera), better cameras, and faster processing would be an almost immediate improvement to current FSD without having to improve software.
There's a lot of people in the world, I'm sure some said that. I would recommend drawing your own conclusions based on it's current state, rather than focusing too much on what people say.
You can check what I was saying a 2 year and 1.5 years ago:
I want from 50% chance after Sep 2024 to 50% chance after Aug 2027.
Rate of improvement IMO seems to be around 2x better every 6 months. But there is still a long long way from driving millions of miles without disengagement to meet their internal benchmark.
My very rough guess is that they still need to make it some 10000x better.
log2(10000) ∗ 0.5 year = 6.64 years
The improvement needed also roughly matches Tesla internal expectations.
According to CJ Moore, Tesla’s director of autopilot software:
> The ratio of driver interaction would need to be in the magnitude of 1 or 2 million miles per driver interaction to move into higher levels of automation.
To be fair I don't think it will be robotaxi level in 2 or 3 years. I agree that robotaxi level is probably more like 2027.
In 2 or 3 years by successful I mean that your average Tesla owner will use it in their every day, and will be able to handle their daily commute fully. But they'll still be behind the steering wheel.
I'm really curious how it will work out for them. The conventional wisdom in self-driving space seems to be that humans are not reliable controllers of automated systems that need input once every day or so.
I'm quite sure that there will be a lot of people overestimating this system if it tries to kill you only once per 1000 kilometers or once per 2 weeks.
Right now it tries to kill you every other drive, so people really pay attention.
I feel like "Safety Score" was the only thing keeping this "FSD" beta to somewhat responsible adults (e.g. risk of losing access). When that they've opened the floodgates, how long before people are reading a book or on their phone while their car "self drives?"
It has a long way to go for sure. But remember that these channels focus on the hard stuff (narrow unmarked roads, unprotected left turns, etc). Also maybe I wasn't clear, but by successful I mean that your average Tesla owner will use it in their every day, and will be able to handle their daily commute fully. But they'll still be behind the steering wheel. I don't expect that in 3 years you'll be able to sleep while it drives.
This is why I immediately came to the comments. My only question was whether this is actually full self driving or if it's just Tesla's usual fraudulent and dangerous (imo) use of the term. I guess it's the latter, thanks.
How can they possibly do this without exposure to liability?
At some point a vehicle is going to do something atrocious, and the owner will have the means (or find a sympathetic litigator) to sue the crap out of the company.
No indication of whether those are autopilot or FSD, or whether the Tesla was at fault.
Musk used to state no FSD deaths during quarterly conference calls, I think he's stopped so there probably has been some.
I don't expect FSD to be 100% perfect, but I want it to be about 10X safer than humans. Any less safe leaves too many unknowns. Any more safe means we waited too long to enable it for unnecessary deaths.
> How can they possibly do this without exposure to liability?
They can’t, just like human drivers can’t drive without liability, which is why insurance is required for all drivers.
Auto insurance companies determine rates based on risk that’s assessed based on many factors related to the driver. They could do the same by assessing the risk associated with using the software. At least then it would have some kind of independent review.
I don't understand why this is legal. First how is it not false advertisement that a feature called "Full Self-Driving" isn't that? This isn't a marketing mistake, it is lying to users about what this technology does.
Second why doesn't Tesla have to submit to some sort of third-party independent verification that this software works? Why does Elon, who has demonstrated over the last month his absolute contempt for his fellow humans, get to decide that I am now an unwilling member of their beta testing as someone who shares the road with Teslas? How dangerous is the software, how often does it make mistakes, what has been done to correct known errors and limitations of relying exclusively on cameras, all of this should be in the public record and independently verified.
What's the logic here? Elon and Tesla, whose entire stock price and financial fortunes are based on this technology working as advertised, are going to pull the plug if the results aren't up to an acceptable level of safety? Why would we think either party would ever do that?
>Second why doesn't Tesla have to submit to some sort of third-party independent verification that this software works
That is the law in the EU but all currently driven Teslas that have been updated in areas that change steering, braking, power, safety features, etc. have the same problem. They haven't been re-approved either and are by definition illegal to drive but rules and laws are there to be ignored after all. I couldn't make changes to my car or motorcycles in those areas if I wanted them to pass a test (like a MOT test) but I'm not a US owned company that you need hundreds of lawyers to even dare look at.
You don't even need to activate FSD to make the car illegal to drive. Just add a few extra HP or upgrade the brakes (hardware or software) and that's it it.
Our current elites do not want lesser humans to make autonomous decisions, hence the push for the technology and the absence of regulations.
We are supposed to be passive, sit in a car while tweeting and be surveilled and tracked 24/7.
(In the case of Tesla, having excellent connections to subsidies, DARPA, as well as providing Starlink Internet to crisis regions with plausible deniability could help, too.)
I don’t understand the demand for “legality”. FSD is like a driver with a fresh license - both are a little dangerous in the beginning but both will get better over time. Also the FSD’s promise (be it Tesla or anyone else) is so far reaching that this “a little dangerous in the beginning“ is worth some elaboration.
Just imagine reading a book in a self driving car and thinking about a new generation of self driving:-)
But that fresh licensed driver is still liable for their mistakes, even deadly ones, is Tesla accepting legal liability for mistakes that occur while "FSD" is enabled? And by that I do also include accidents and injuries that occurred even if "FSD" shuts itself off a few seconds before it hits the diver/bollard/stopped fire truck/etc.
> I don’t understand the demand for “legality”. FSD is like a driver with a fresh license - both are a little dangerous in the beginning but both will get better over time.
Liability is the answer here, the driver is liable and so needs to adjust their risk-taking because of possible consequences. FSD failures has no consequences to the responsible party (Tesla and other automakers, etc.) if legality isn't in question.
> Also the FSD’s promise (be it Tesla or anyone else) is so far reaching that this “a little dangerous in the beginning“ is worth some elaboration.
What do you mean worth some elaboration? So elaborate on it, not sure if you stated what you meant. Even then, I'm a user of the road and I've not been asked to participate in Tesla's or any automakers experiments, this is a public good, used by real people everyday, not some laboratory to be experimented on and tweak your experiment accordingly.
If someone you love is killed by a FSD failure in public roads, would you still defend it? And I know that appealing to this emotional side might sound like an exaggeration but this is what might happen to someone else, someone they love might be killed by this experiment, are you willing to accept that damage to be done to your family in name of progress of a proprietary technology being developed for profit?
> Just imagine reading a book in a self driving car and thinking about a new generation of self driving:-)
I don't want to be in a self-driving car reading a book, why would I need that if I have access to trains carrying me faster than any car could on a road and when I can hail a cab for the last stretch if that isn't served by any mass transportation system? Things that aren't just existing technology but also much more environmentally friendly than having 1 car per person or household?
I dream of a world with less need of cars, with less cars taking space in city centres. Cars brought us far but also has shown us that aren't sustainable for the longer term (next centuries), I don't dream about having a car that drives me around, it sounds quite dystopian to be honest.
The promise of future cars is a traffic without any dependence on human errors. For that I am ok with an inevitable transition which may bring some serious accidents.
The problem with trains is to get into one. A car can be boarded almost anywhere. And if that car doesn’t need any human operator – it’s perfect.
Just imagine all the people who will not need to drive. Machines will drive and people will design yet another better future:-)
A little off-topic, but: I just found out about Comma.ai and... is it really legal in the US? I know it wouldn't be in most (all?) EU countries, as modifications to the vehicle require certification by a professional engineer.
It's kind of terrifying that anyone can modify their vehicle as they please and drive it on the road with everyone else.
This will kill and injure people in persuit of at least one of Tesla's profits or technological progress. It's interesting to see the USA still on the side of reckless with human life in favour of progress. I don't believe Europe would take the same stance, which I suppose is a balance between short and long term thinking.
Attentive subset of good drivers doing millions of miles vs general public is a big difference. Is FSD's reputation being saved by the small pool of good drivers rather the reverse? If so, worst case scenario will be pretty grim.
Currently at rank #348 (page 12), seems odd given its points, freshness and number of comments.
Curiously on page 12, most posts have 1 or no comments. Only one other post has more comments - "The Twitter Advertiser Exodus" [1] clocks in with 315 although it is already a day old.
92 comments
[ 4.3 ms ] story [ 163 ms ] threadIt's a commercial name for a Level 3 autonomous function on Tesla cars ("the car’s systems perform the entire dynamic driving task (DDT) within the area that it is designed to do so. The human driver is only expected to be responsible for the DDT-fallback when the car essentially “asks” the driver to take over if something goes wrong or the car is about to leave the zone where it is able to operate." [1]), considering that the "area" is not strictly defined, and could be any road with sufficient visual cues for the car to assume where a viable lane could be.
In terms of sensors, it relies exclusively on optical cameras, as opposed to other autonomy projects that use LiDAR and/or radar and/or ultrasounds as well as cameras.
People have been paying thousands of dollars to subscribe to FSD for many years now, even when the only autonomous interventions in the car were strict Level 2 autonomy. Contributing factors for this seemingly irrational investment could be that the Level 2 features worked great (except for phantom braking and other arguably dangerous mistakes), and the backhanded marketing that would place FSD as a Level 5 autonomy feature.
[1] https://level5auto.com/what-is-level-3-autonomous-driving/
SAE automation level 2 means the machine can do some of the boring parts of driving, but you as driver are actually always responsible, you are required to overrule the machine whenever it does something which was a bad idea and that's somehow not Tesla's fault. If you somehow fall asleep while a level 2 car is "driving" you're possibly never going to wake up, and if you do chances are any collision or other damage is your fault.
At level 4, the machine will either do its job or safely give up. So it's OK to be asleep, although it's possible when you wake up you're like "God damn it car, I wanted to be in Boston by now, and you've just sat here moaning about a little bit of snow just outside Plattsburgh all night"...
(At level 5 the machine should drive in any reasonable conditions, this probably gets very elastic about what constitutes a "reasonable" condition for driving - is this car level 5 despite the fact it won't drive in a blizzard? Maybe not driving in a blizzard isn't cowardice but common sense? A dirt track seems fine in a 4WD with AI, but how about this boulder-strewn mountain path?)
My guess is that FSD will be in Beta until such metric is defined, measured, and indeed FSD comes ahead of humans.
130,000 Cyclists are injured.
https://www.iihs.org/topics/fatality-statistics/detail/pedes...
https://www.cdc.gov/transportationsafety/bicycle/index.html
We shouldn’t be accepting any deaths, or accidents, and take steps to reduce (eliminate) them.
However, people seem to really get worried when we want technology to assist us.
About 18 pedestrian and 2-3 cyclists will die today in the US alone. What should we do?
The humans are careless, often taking their eyes off the road. I see people looking at devices in their hands while driving.
People bike near me on curvy roads and people drive fast.
Hear about these horrible accidents where cars collide with cyclists.
I think humans could use a little more assistance.
We haven’t even discussed the 38,000 traffic deaths in the US alone every year.
Is this acceptable?
The dude behind the mannequin video sells lidar systems and was caught doctoring the first video he released.
https://twitter.com/RealDanODowd/status/1559245760054575112?...
What about the one from China where a Tesla accelerated on its own, the driver was unable to brake, and it ended up mowing down two pedestrians?
https://electrek.co/2022/11/13/tesla-china-responds-to-drama...
2) lidar guy. You'll see a ton of responses to his antics on HN and YouTube.
3) not fsd, driver confused pedals
Says Tesla who had every intention to lie. The driver wasn’t drunk and I can’t believe a human can plough accudentally for that long.
Not that this is relevant anyway. This incident is not related to autonomy at all.
So you're saying we should trust full self-driving on a car that can't brake in emergencies reliably? And that this is ok?
>2) lidar guy. You'll see a ton of responses to his antics on HN and YouTube.
From fans of Tesla & Musk. Just as untrustworthy. Except one side here has provided evidence of a problem, and the other side has responded by making excuses & attacking the guy.
Basically, FSD has virtually no safety record, not a good safety record.
2 years ago, FSD was pretty awful. Back then, my opinion was that Tesla FSD was doomed: Self driving without LIDAR was not going to be feasible for at least a decade. Now, I've changed my mind. Current version of course has a long way to go, but the current state and the rate of progress makes me believe now that, in 2 or 3 years, with a new hardware update, Tesla FSD will be very successful.
I recommend checking out some videos of the latest version (10.69.3.1) to get an idea of how good (or bad) it currently is.
[1] https://www.youtube.com/@ChuckCook https://www.youtube.com/@AIDRIVR
Do you think that folks who bought FSD will get a hardware update, without extra cost?
Camera placement (A-pillar camera), better cameras, and faster processing would be an almost immediate improvement to current FSD without having to improve software.
Isn't that exactly what people were saying 2 or 3 years ago?!
I want from 50% chance after Sep 2024 to 50% chance after Aug 2027.
Rate of improvement IMO seems to be around 2x better every 6 months. But there is still a long long way from driving millions of miles without disengagement to meet their internal benchmark.
https://www.metaculus.com/questions/11608/self-driving-taxis...
Full perfect self driving was complete and ready for release in 2017
They seem to be going at about 2x improvement rate every 6 months.
https://www.metaculus.com/questions/11608/self-driving-taxis...
My very rough guess is that they still need to make it some 10000x better.
log2(10000) ∗ 0.5 year = 6.64 years
The improvement needed also roughly matches Tesla internal expectations.
According to CJ Moore, Tesla’s director of autopilot software:
> The ratio of driver interaction would need to be in the magnitude of 1 or 2 million miles per driver interaction to move into higher levels of automation.
So, my current prediction is Aug 2027:
- 5% before July 2023
- 25% before Aug 2025
- 50% before Aug 2027
- 75% before May 2030
- 95% before Apr 2034
In 2 or 3 years by successful I mean that your average Tesla owner will use it in their every day, and will be able to handle their daily commute fully. But they'll still be behind the steering wheel.
I'm quite sure that there will be a lot of people overestimating this system if it tries to kill you only once per 1000 kilometers or once per 2 weeks.
Right now it tries to kill you every other drive, so people really pay attention.
https://www.youtube.com/watch?v=zoKAX-F1t9U
I feel like "Safety Score" was the only thing keeping this "FSD" beta to somewhat responsible adults (e.g. risk of losing access). When that they've opened the floodgates, how long before people are reading a book or on their phone while their car "self drives?"
Here's something more resembling an average drive from that same channel https://www.youtube.com/watch?v=b_1NOiFKukg
Yikes.. works something like 100%, 60% of the time?
I think I'll wait for the "Full Self-Driving Pro MAX" version when it comes out. ( If I don't get runover by a Beta version in the mean time.. )
[1] https://www.motor1.com/news/575167/mercedes-accepts-liabilit...
At some point a vehicle is going to do something atrocious, and the owner will have the means (or find a sympathetic litigator) to sue the crap out of the company.
Multiple people have already died[0]
[0] https://fortune.com/2022/10/19/tesla-cars-involved-in-10-of-...
Musk used to state no FSD deaths during quarterly conference calls, I think he's stopped so there probably has been some.
I don't expect FSD to be 100% perfect, but I want it to be about 10X safer than humans. Any less safe leaves too many unknowns. Any more safe means we waited too long to enable it for unnecessary deaths.
-- Tesla owner without FSD.
They can’t, just like human drivers can’t drive without liability, which is why insurance is required for all drivers.
Auto insurance companies determine rates based on risk that’s assessed based on many factors related to the driver. They could do the same by assessing the risk associated with using the software. At least then it would have some kind of independent review.
Second why doesn't Tesla have to submit to some sort of third-party independent verification that this software works? Why does Elon, who has demonstrated over the last month his absolute contempt for his fellow humans, get to decide that I am now an unwilling member of their beta testing as someone who shares the road with Teslas? How dangerous is the software, how often does it make mistakes, what has been done to correct known errors and limitations of relying exclusively on cameras, all of this should be in the public record and independently verified.
What's the logic here? Elon and Tesla, whose entire stock price and financial fortunes are based on this technology working as advertised, are going to pull the plug if the results aren't up to an acceptable level of safety? Why would we think either party would ever do that?
That is the law in the EU but all currently driven Teslas that have been updated in areas that change steering, braking, power, safety features, etc. have the same problem. They haven't been re-approved either and are by definition illegal to drive but rules and laws are there to be ignored after all. I couldn't make changes to my car or motorcycles in those areas if I wanted them to pass a test (like a MOT test) but I'm not a US owned company that you need hundreds of lawyers to even dare look at.
You don't even need to activate FSD to make the car illegal to drive. Just add a few extra HP or upgrade the brakes (hardware or software) and that's it it.
Our current elites do not want lesser humans to make autonomous decisions, hence the push for the technology and the absence of regulations.
We are supposed to be passive, sit in a car while tweeting and be surveilled and tracked 24/7.
(In the case of Tesla, having excellent connections to subsidies, DARPA, as well as providing Starlink Internet to crisis regions with plausible deniability could help, too.)
There is no way that they could release it so easily in EU, but this is also why all startups and big tech companies are in USA.
Just imagine reading a book in a self driving car and thinking about a new generation of self driving:-)
There also needs to be someone who watches over them :) .
Liability is the answer here, the driver is liable and so needs to adjust their risk-taking because of possible consequences. FSD failures has no consequences to the responsible party (Tesla and other automakers, etc.) if legality isn't in question.
> Also the FSD’s promise (be it Tesla or anyone else) is so far reaching that this “a little dangerous in the beginning“ is worth some elaboration.
What do you mean worth some elaboration? So elaborate on it, not sure if you stated what you meant. Even then, I'm a user of the road and I've not been asked to participate in Tesla's or any automakers experiments, this is a public good, used by real people everyday, not some laboratory to be experimented on and tweak your experiment accordingly.
If someone you love is killed by a FSD failure in public roads, would you still defend it? And I know that appealing to this emotional side might sound like an exaggeration but this is what might happen to someone else, someone they love might be killed by this experiment, are you willing to accept that damage to be done to your family in name of progress of a proprietary technology being developed for profit?
> Just imagine reading a book in a self driving car and thinking about a new generation of self driving:-)
I don't want to be in a self-driving car reading a book, why would I need that if I have access to trains carrying me faster than any car could on a road and when I can hail a cab for the last stretch if that isn't served by any mass transportation system? Things that aren't just existing technology but also much more environmentally friendly than having 1 car per person or household?
I dream of a world with less need of cars, with less cars taking space in city centres. Cars brought us far but also has shown us that aren't sustainable for the longer term (next centuries), I don't dream about having a car that drives me around, it sounds quite dystopian to be honest.
The problem with trains is to get into one. A car can be boarded almost anywhere. And if that car doesn’t need any human operator – it’s perfect.
Just imagine all the people who will not need to drive. Machines will drive and people will design yet another better future:-)
It's kind of terrifying that anyone can modify their vehicle as they please and drive it on the road with everyone else.
https://insideevs.com/news/603364/tesla-probe-motorcyclist-c...
Curiously on page 12, most posts have 1 or no comments. Only one other post has more comments - "The Twitter Advertiser Exodus" [1] clocks in with 315 although it is already a day old.
[1] https://news.ycombinator.com/item?id=33713142