The article is hysterical clickbait nonsense and loaded with some very basic misunderstandings—seemingly including a fundamental confusion about what "beta" means. Even the headline is flat out false, because the FSD beta software contains numerous "safeguards".
Tesla has deployed this beta software to just 1000 customers who have demonstrated competence and level-headedness. The drivers are required to keep their hands on the wheel, the same as with the adaptive cruise / lane centering features available on many new cars.
Yes, and Tesla also made sure that all the other road users around those 1000 beta testers opted in to being part of a development project. So all good.
I've lived in Paris for 15+ years, where many drivers are worst than FSD. I've watched hours of beta on Youtube. I never opted in, either.
Many human drivers don't care about cyclists safety: they overtake at high speed, sometimes within centimeters, and occasionally force us to change course or hit us (and don't stop to help unless we don't get up).
Yet, we don't make such drivers take an exam to check their skills/behavior because they passed one many years ago.
I'd like to know how many accidents have happened with FSD before we slow down self-driving (instead of the opinion of consumer groups who've always sided against this technology). Otherwise, we should put dangerous drivers out of the road _now_. We rarely do, even after they hit pedestrians or cyclists while texting, so I prefer we focus on improving driver assist systems and focus on actual accidents (instead of Tesla brushing against bushes).
I did! Multiple times, including one where I was hit from behind at high speed, injured, the driver stopped and left without helping. I file a claim ("contre X") at the police station and gave them the license plate (which they also heard when I called 911 right after it happened).
This went nowhere, because the police was overwhelmed (that was during the gilets jaunes movement, and the police was busy going after people for destruction of public properties... priority, priority...)
You drive around every day on roads with learners, unroadworthy vehicles with bald tyres, and the occasional angry psycho driving a stolen car. You didn't opt into any of that either.
That wasn't an argument defending Tesla or an argument for what ought to be illegal. It was merely an argument against the silly notion that other drivers on the road didn't "opt in" to this. Of course they didn't "opt in"—because that's not how roads work. You share the road with a huge variety of different vehicles, being controlled in a huge variety of different ways. There's no "opt in" for any of it.
>Tesla has deployed this beta software to just 1000 customers
Source? As far as I know, thousands of Tesla employees are in the beta but only dozens of non-Employees have been enrolled to the program and had one on one conversations with Tesla employees.
For several years now, in the EU at least, all new cars need to have automatic emergency breaks. It's basically impossible to run into something, run someone over, etc. This kind of system reduces stress and gives at least me a certain peace of mind. In this regard Tesla self driving is a step backwards.
You seem to be confused about what AEB is and what it does. It is generally effective at braking for static or rapidly slowing vehicles ahead. Most AEB systems are not effective "safeguards against driving into oncoming traffic and parked cars". These would involve a steering intervention, not a braking intervention.
I assume AEB ignores a lot of stuff on purpose, otherwise it'd be braking due to things on the side of the road, and cars in the other lane on a curve that are heading straight towards you
Correct. Most AEB systems use a forward facing radar, which often receive spurious reflections from various street elements overhead bridges, overhead signage and manhole covers. As a result, most radar-based implementations will actively ignore completely stationary objects while driving at high speed.
That is a good article which summarises the state of AEB and its general capability. It's worth noting that the most effective system they tested doesn't use radar at all, it was an entirely vision-based system: Subaru Eyesight. Worth remembering this when people agonise over Tesla ditching radar sensors on the Model 3.
I chose my parents' most recent new car precisely because of this—a 2019 model year Subaru Forester. Not only was the active safety top notch, its chassis tuning means that it's more capable of remaining composed after swerving to avoid an obstacle. (It was also one of the few vehicles which ticked every box for them.)
Here is the comparison I relied upon for that decision. (Yes, I am Australian.)
Well,the rest have radar, so that's not a problem.
"Stereoscopic vision works most effectively for distances up to 18 feet. Beyond this distance, your brain starts using relative size and motion to determine depth."
If you're interested, Tesla's head of AI recently did a public presentation of their new vision based depth system. It's worth a watch if you find this stuff interesting and enjoy learning about the forefront of technology.
A key takeaway is that they've been running this stack in shadow mode (validating output but not controlling the car) on everyone's Tesla for quite some time. Equivalent to 1000 YEARS worth of real world driving. And from this data they've proven it is now superior to radar in all circumstances.
being very suspicious about that talk, the output they show doesn't matches at all with the output people have extracted from running teslas - https://twitter.com/greentheonly/status/1412597377228226562 - specifically the 'trained' heath signature running horizontally across the dash.
Wait, yes they do. Every other car has a driver who, presumably, does not want to randomly veer off into concrete pillars or oncoming traffic like FSD seems to do with an alarming frequency. Humans are imperfect and make mistakes when distracted or driving illegally (under the influence of alcohol or drugs or without a license or with poor eyesight...) but the argument that a Tesla operating under FSD is just as safe as this most dangerous class of drivers is an extremely unconvincing one.
There are human drivers in Tesla vehicles too. And FSD beta isn’t doing anything fundamentally different than any other “regular” car with adaptive cruise and lane centring—it is steering, braking and accelerating based on conditions reported by sensors. The only difference is that the FSD beta is capable of doing this task in a wider variety of scenarios.
Why call it fsd when it can’t do fsd? Also if I need to watch my car driving, then why should I pay $199 a month for it rather than drive the car myself, or call an Uber.
I'm not defending the FSD name. Personally I think the name is stupid.
I'm not defending the price. Personally I wouldn't pay for it.
For the record, I don't own a Tesla vehicle. I don't know anyone who owns a Tesla vehicle. I don't own TSLA stock (or any other individual stock or derivatives). I've never driven a Tesla vehicle or even sat inside any Tesla vehicle.
When does it cross the line from "bad naming" to "fraud"? If I call my product a "life jacket" and you wear it while out on the water but all it does is make you sink faster it's not just a bad name.
So long as Tesla correctly explains what FSD does to its paying customers, then it is not fraud. It is inconceivable that anyone could go through the Tesla buying process and check the box for "FSD" without reading the clear description of what the feature actually is. This isn't like installing iTunes, it adds $8k to the purchase price of the car and it's not optioned by default.
It's true that some people who haven't purchased a Tesla vehicle might be misinformed about FSD. But it's not fraud if they aren't customers.
I don't like the FSD name and I think Tesla should change it. But I can see how people can come to different conclusions about it. The word "full" can mean very different things depending on how you think about the word and the context of its use. The two extremes:
MAXIMALLY NARROW — "Full" self driving means the car can reliably control the vehicle perfectly in all circumstances, on all road surfaces, at all times, in all conditions. This is something that most HUMANS are incapable of, by the way. So under this definition, most humans aren't "full" drivers either.
MAXIMALLY BROAD — "Full" self driving can perform the "full" set of driving tasks, imperfectly, in certain circumstances, on some road types, only at some times, only in some conditions. Doesn't even mean it can drive in these limited situations well, only that it can drive them.
Most people would agree that a realistic definition fits somewhere between those extremes. But there's no question that FSD satisfies the maximally broad definition.
In every car on the road, it is possible for the driver to take their hands completely off the wheel. The only variable between cars is how long until there are consequences. In most cars, the consequences would be somewhere between immediate and imminent.
In the case of any vehicle with adaptive cruise control and lane centring features, this could be anywhere from many seconds to many minutes.
In the case of a Tesla with Autopilot engaged, no hands on the steering wheel for 30 seconds will—if there haven't been any consequences prior—result in the car turning on the hazards, slow down and then stop.
I need to come up with a snappy name for this, but I judge progress of things by the level of complaints. It seems the media isn't really set up to say, "yeah on balance this is probably better (or soon to become so) on average" as that's not attention grabbing.
However, imagine finding yourself in some unlikely time travel event and seeing a headline from 5 or 10 years in the future and it's talking about how a car that drives itself brushed against some bushes as the major complaint.
Musk said that their cars would be able to drive without human intervention coast to coast by 2017, that's nearly 5 years ago and it can't even navigate across a single city. At this rate it's more like 30-50 years away.
If your grandfather drove bad enough to pull two such stunts on a single drive he wouldn't have a car to drive anymore. So an alpha flag would at least indicate: "drives literally worse than rcMgD2BwE72F's grandfather" .
Erratum: my grandpa drives worst than FSD with human oversight (like in all the video you can find on Youtube). I didn't meant FSD without oversight, which is not possible anyway. But are we discussing the current level of FSD as a Level 5 system? Don't think so.
why is it so hard to understand that roads full of drivers equal to your mythical grandpa would be a really bad idea and people would be dying left and right?
>my grandpa drives worst than FSD with human oversight
Your grandpa is one person. Retesting 100 million drivers to find the bottom 10% is not exactly very feasible. For a model of FSD, all vehicles on the road share the exact same level of performance. We are talking about scale here.
It is like comparing me selling some vegetables from my garden to a neighbor and a company selling 100 million units of vegetables across a country. It makes zero sense to say "they should both have the same level of safety standards", since a listeria outbreak sickening thousands is far worse than a tainted cucumber poisoning one neighbor.
Tens of thousands of cars using dangerously poor driving level software is worse than any one individual driver that may have dangerously poor skills.
In the U.S. they expire but you're usually not retested unless you move states and sometimes not even then. You usually just go down to the state's motor vehicle office and confirm your info is correct, pay the renewal and they print out a new license
Yep, the extent to which people are allowed to renew their license in the U.S. is very nearly criminal (both in the cases of old age and in previous history of driving impaired or recklessly). It's one of the less talked about externalities of the States being a car-based culture. Here, driving is considered a right more than a privilege, whereas in most other first world countries (e.g. most of Europe) the opposite is true.
If his grandpa really drives this poorly he will lose his ability to drive well before it expires either directly from accidents, from losing his license after accidents, or from his insurance becoming too expensive. I'm always amazed at how badly die-hard FSD supporters think humans drive
Your grandfather's license should be taken away (and, if he drives anyway, his car should be confiscated so that he stops before more drastic solutions are needed)
One problem we have here (in the UK) is that unless you've committed a crime the most likely person to assess that you shouldn't drive any more is your GP (doctor). But doctors operate in an ethical framework that prioritises the interests of each patient as an individual, resolving both silly theoretical ethical problems ("Do I kill this healthy patient to save these other dying patients?") and real world practical ones. So, "grandpa's driving is bad" only meets the threshold when it's so bad he'd likely die, if he will just kill or injure other people that's not an ethical obstacle in this framework unless grandpa is uncomfortable with it, in which case he should stop driving without waiting for the license to be taken away.
This is something you're seeing in health policy right now. Sticking needles in people's arms is something medics do, but they all operate under this ethical framework. So, while a government may have good reason to vaccinate Bob, the doctor is only willing to vaccinate Bob if Bob wants to be vaccinated and it's personally in Bob's interest, regardless of whether it'd be better for everybody else.
Still, where it's important to public policy you can sidestep the ethical problem by sidestepping the medical procedure, here are two examples used today in the UK:
1. If somebody claims you're the other parent of their children and you owe child support but you disagree, the government can decide to make you get a DNA test (you get a refund if they aren't your kids). But the DNA test is a medical procedure, so if you really don't want it, the medics won't do it. Instead you'll end up in court, and the court will look at the available evidence and most likely say, well, seems to us if you were not the parent of these kids you'd have taken that DNA test and proved it. You owe child support. (Owing child support isn't a crime, so, there's no need to meet the criminal evidence standard).
2. If you're in a vehicle accident or just it seems obvious to them you might be drunk, the police can decide to test for alcohol. Some of the tests they could try don't need a medic but do need your co-operation, others would need a medic. Either way you might refuse and so, the law just provides for a penalty for "Refusing to provide a sample" that's almost exactly the same as the penalty for drink driving you might hope to avoid by doing so.
I agree that people skills change over time, which is why I, as a driver, am in favor that every driver should retake the driving exam and a health exam every 2 years.
I am also in favor of enforcing this for all foreign drivers going through my country somehow, since most of them don’t really know how to drive here.
In 8:39 - 8:42, there's a group (three people with a dog and stroller) crossing the street left to right.
Here's what the Tesla recognises on the screen:
One person
Nothing
Motorcycle
Nothing
Motorcycle
Nothing
Car
Nothing
One person
Two persons
One person
One person and a motorcycle
One person
Nothing
Motorcycle
One person
One person and two motorcycles
Two persons
One person
One person and a motorcycle
Two persons
One person
It seems to me you wouldn't really need public testing yet at this early stage of development. At a minimum, I would expect it to count the people correctly and get rid of the motorcycles, i.e. basic perception.
Absolutely, an essential part of driving is predicting how the state of the world will change in the future - and the behaviour of a motorcycle is vastly different from the behaviour of a stroller.
In this case it doesn't matter, because there's a car in front anyway.
I don't know how much the system can ignore or mislabel people and still drive safely, but right now it's not really safe and it's not necessary to have the car drive while improving object recognition.
I think self driving has no future because it concentrates defendants into single juicy target.
Normal cars are random Joe vs. random Jane, both probably poor, lawsuit is not viable. Self driving cars is everybody vs. rich company. It's worth a try.
Counterpoint: whoever can convince insurance companies that they can reduce car accidents significantly might be in a very good position to strike a nice deal with one or more insurance companies.
I watched the video linked, that software should not even be rated 'beginner driver' and nearly caused several accidents in the (short) video. What bothers me most is that there were several pedestrians clearly visible that it missed completely on the UI, it does a lot better with other vehicles. It also does not seem to correctly identify people on shared mobility devices (a couple of those electric scooters appear).
Exactly. Seeing is not understanding, and I take from objects popping in and out of existence that the software will be happy to erase you from its reality once confidence drops below a certain threshold. It does so with vehicles as well (also stationary ones), but those are presumably a lot more solid than those 'drive through pedestrians'.
I mean it's not showing to the user everything it's using to route through the world, you can't tell from that view what it's going to run straight through
That argument works both ways: you also can't tell from that view what it will stop for (or run straight through). If it is just infotainment then they should switch it off.
Meanwhile, you can't walk two blocks in SF without seeing a Cruise or Waymo car, and those are driving thousands of miles between disengagments. This looks like a hobby project in comparison.
If not for the reactionspeed of the guy making that video there were two instances already, one where the car suddenly swerved right into another car, another where it aimed for a concrete pillar.
Seriously, watch the video. Human not wanting to die is the default, but should not be a requirement for something labelled 'full self driving'.
I watched it a day ago. It's not labelled full self driving, it's labelled full self driving beta "warning, the car may do the worst thing at the worst time, be alert"
What would be useful is to get the debug report, what it thought it was doing, what it was planning to do in the next second, would it have crashed?
I want a law passed that requires the installation of a giant light on top of the car to indicate to me that it is in self-driving mode and may do something unpredictable at any point. At least give me a chance of avoiding a crash, especially if I am underway on a motorbike.
As a pedestrian, I can make eye contact with a driver before crossing the street. I can’t do that with a self-driving car. I just have to hope the software is bug free and the various sensors are accurately detecting my presence.
"but what about human drivers" is not a constructive argument here. I agree with the parent question because I'm worried FSD will suddenly believe it's in a slalom competition on a straight road and hit me when we're parallel to eachother. That's something FSD is likely to do today that I don't expect any sober human driver to do.
There are approximately 1000 vehicles on the road which are equipped with the Tesla FSD Beta software. To date, no vehicles equipped with this FSD Beta have been involved in any collision while the software has been (or recently had been) in control.
They’ve doubled it and likely will continuing rolling it out. Thankfully, as you said, none have been involved in an accident.
If you look at YouTube there are tons of sleeping Tesla drivers on the highway and I think a lot are just pranks. Once people start doing stupid stuff like this on normal roads it’s not going to be fun for the people on the receiving end.
103 comments
[ 3.5 ms ] story [ 175 ms ] thread- "Videos of FSD beta 9 in action don’t show a system that makes driving safer or even less stressful"
- "Tesla should at the very least be monitoring drivers in real time to ensure that they’re paying attention while using new software"
They do https://www.cnbc.com/2021/05/28/tesla-starts-using-cabin-cam...
https://twitter.com/greentheonly/status/1312607693249667073
Tesla has deployed this beta software to just 1000 customers who have demonstrated competence and level-headedness. The drivers are required to keep their hands on the wheel, the same as with the adaptive cruise / lane centering features available on many new cars.
Many human drivers don't care about cyclists safety: they overtake at high speed, sometimes within centimeters, and occasionally force us to change course or hit us (and don't stop to help unless we don't get up).
Yet, we don't make such drivers take an exam to check their skills/behavior because they passed one many years ago.
I'd like to know how many accidents have happened with FSD before we slow down self-driving (instead of the opinion of consumer groups who've always sided against this technology). Otherwise, we should put dangerous drivers out of the road _now_. We rarely do, even after they hit pedestrians or cyclists while texting, so I prefer we focus on improving driver assist systems and focus on actual accidents (instead of Tesla brushing against bushes).
This went nowhere, because the police was overwhelmed (that was during the gilets jaunes movement, and the police was busy going after people for destruction of public properties... priority, priority...)
Source? As far as I know, thousands of Tesla employees are in the beta but only dozens of non-Employees have been enrolled to the program and had one on one conversations with Tesla employees.
I chose my parents' most recent new car precisely because of this—a 2019 model year Subaru Forester. Not only was the active safety top notch, its chassis tuning means that it's more capable of remaining composed after swerving to avoid an obstacle. (It was also one of the few vehicles which ticked every box for them.)
Here is the comparison I relied upon for that decision. (Yes, I am Australian.)
https://www.youtube.com/watch?v=cs7PSpIMJYk
"Stereoscopic vision works most effectively for distances up to 18 feet. Beyond this distance, your brain starts using relative size and motion to determine depth."
https://www.youtube.com/watch?v=g6bOwQdCJrc
A key takeaway is that they've been running this stack in shadow mode (validating output but not controlling the car) on everyone's Tesla for quite some time. Equivalent to 1000 YEARS worth of real world driving. And from this data they've proven it is now superior to radar in all circumstances.
Too many Tesla apologists around.
I'm not defending the price. Personally I wouldn't pay for it.
For the record, I don't own a Tesla vehicle. I don't know anyone who owns a Tesla vehicle. I don't own TSLA stock (or any other individual stock or derivatives). I've never driven a Tesla vehicle or even sat inside any Tesla vehicle.
It's true that some people who haven't purchased a Tesla vehicle might be misinformed about FSD. But it's not fraud if they aren't customers.
I don't like the FSD name and I think Tesla should change it. But I can see how people can come to different conclusions about it. The word "full" can mean very different things depending on how you think about the word and the context of its use. The two extremes:
MAXIMALLY NARROW — "Full" self driving means the car can reliably control the vehicle perfectly in all circumstances, on all road surfaces, at all times, in all conditions. This is something that most HUMANS are incapable of, by the way. So under this definition, most humans aren't "full" drivers either.
MAXIMALLY BROAD — "Full" self driving can perform the "full" set of driving tasks, imperfectly, in certain circumstances, on some road types, only at some times, only in some conditions. Doesn't even mean it can drive in these limited situations well, only that it can drive them.
Most people would agree that a realistic definition fits somewhere between those extremes. But there's no question that FSD satisfies the maximally broad definition.
They call it FSD in the US because nobody has won a lawsuit against them yet. Doesn’t the US have customer protection agencies ?
In the case of any vehicle with adaptive cruise control and lane centring features, this could be anywhere from many seconds to many minutes.
In the case of a Tesla with Autopilot engaged, no hands on the steering wheel for 30 seconds will—if there haven't been any consequences prior—result in the car turning on the hazards, slow down and then stop.
However, imagine finding yourself in some unlikely time travel event and seeing a headline from 5 or 10 years in the future and it's talking about how a car that drives itself brushed against some bushes as the major complaint.
https://youtu.be/GlIdu7prsAw?t=334
https://youtu.be/GlIdu7prsAw?t=164
I've skipped to spots where the car makes an error, but the whole video is worth a watch.
What the alpha flag would change, exactly?
https://en.wikipedia.org/wiki/Self-driving_car#SAE_Classific...
Your grandpa is one person. Retesting 100 million drivers to find the bottom 10% is not exactly very feasible. For a model of FSD, all vehicles on the road share the exact same level of performance. We are talking about scale here.
It is like comparing me selling some vegetables from my garden to a neighbor and a company selling 100 million units of vegetables across a country. It makes zero sense to say "they should both have the same level of safety standards", since a listeria outbreak sickening thousands is far worse than a tainted cucumber poisoning one neighbor.
Tens of thousands of cars using dangerously poor driving level software is worse than any one individual driver that may have dangerously poor skills.
One problem we have here (in the UK) is that unless you've committed a crime the most likely person to assess that you shouldn't drive any more is your GP (doctor). But doctors operate in an ethical framework that prioritises the interests of each patient as an individual, resolving both silly theoretical ethical problems ("Do I kill this healthy patient to save these other dying patients?") and real world practical ones. So, "grandpa's driving is bad" only meets the threshold when it's so bad he'd likely die, if he will just kill or injure other people that's not an ethical obstacle in this framework unless grandpa is uncomfortable with it, in which case he should stop driving without waiting for the license to be taken away.
This is something you're seeing in health policy right now. Sticking needles in people's arms is something medics do, but they all operate under this ethical framework. So, while a government may have good reason to vaccinate Bob, the doctor is only willing to vaccinate Bob if Bob wants to be vaccinated and it's personally in Bob's interest, regardless of whether it'd be better for everybody else.
Still, where it's important to public policy you can sidestep the ethical problem by sidestepping the medical procedure, here are two examples used today in the UK:
1. If somebody claims you're the other parent of their children and you owe child support but you disagree, the government can decide to make you get a DNA test (you get a refund if they aren't your kids). But the DNA test is a medical procedure, so if you really don't want it, the medics won't do it. Instead you'll end up in court, and the court will look at the available evidence and most likely say, well, seems to us if you were not the parent of these kids you'd have taken that DNA test and proved it. You owe child support. (Owing child support isn't a crime, so, there's no need to meet the criminal evidence standard).
2. If you're in a vehicle accident or just it seems obvious to them you might be drunk, the police can decide to test for alcohol. Some of the tests they could try don't need a medic but do need your co-operation, others would need a medic. Either way you might refuse and so, the law just provides for a penalty for "Refusing to provide a sample" that's almost exactly the same as the penalty for drink driving you might hope to avoid by doing so.
I am also in favor of enforcing this for all foreign drivers going through my country somehow, since most of them don’t really know how to drive here.
Here's what the Tesla recognises on the screen:
It seems to me you wouldn't really need public testing yet at this early stage of development. At a minimum, I would expect it to count the people correctly and get rid of the motorcycles, i.e. basic perception.I don't know how much the system can ignore or mislabel people and still drive safely, but right now it's not really safe and it's not necessary to have the car drive while improving object recognition.
Normal cars are random Joe vs. random Jane, both probably poor, lawsuit is not viable. Self driving cars is everybody vs. rich company. It's worth a try.
Counterpoint: whoever can convince insurance companies that they can reduce car accidents significantly might be in a very good position to strike a nice deal with one or more insurance companies.
This crap should not be in control of a vehicle.
It's beta, why would you expect the part where it displays stuff for humans to be perfect?
v8 showed more info, v9 is fancier, but slightly less useful visualizations for now.
If it tries to drive into someone who vanished from the screen I'm sure we'll hear about it.
And release version HumanDrivers who kill 3000 people per day.
It seems to cope find with moderately busy streets https://www.youtube.com/watch?v=EtPGTOJ5Nks
I live in the examination territory for my region, in a country where you need to do some serious learning before you're allowed on the road in a car.
Have any FSD beta Teslas crashed?
Seriously, watch the video. Human not wanting to die is the default, but should not be a requirement for something labelled 'full self driving'.
What would be useful is to get the debug report, what it thought it was doing, what it was planning to do in the next second, would it have crashed?
I think that's a good assumption for most human driven cars as well :)
Has anyone sued Tesla for being hit by their self driving tech?
If you look at YouTube there are tons of sleeping Tesla drivers on the highway and I think a lot are just pranks. Once people start doing stupid stuff like this on normal roads it’s not going to be fun for the people on the receiving end.
https://www.torquenews.com/11826/elon-musk-tesla-fsd-beta-te...
I haven't seen a single one that didn't obviously look like a prank.
I’d would Lose. My. Mind.
Maybe these cities could furthermore be child-free, a la GTA V, so only consenting adults are subject to these heightened risks.