Tesla Accused of Deception in Promoting Autopilot
bloomberg.com
bloomberg.com
"Build upon Enhanced Autopilot and order Full Self-Driving Capability on your Tesla. ... All you will need to do is get in and tell your car where to go. If you don’t say anything, the car will look at your calendar and take you there as the assumed destination or just home if nothing is on the calendar. Your Tesla will figure out the optimal route, navigate urban streets (even without lane markings), manage complex intersections with traffic lights, stop signs and roundabouts, and handle densely packed freeways with cars moving at high speed. When you arrive at your destination, simply step out at the entrance and your car will enter park seek mode, automatically search for a spot and park itself. A tap on your phone summons it back to you."
Tesla does not have that technology. Not even close. Their cars do not have the sensor suite to do it. They did a demo back in 2016 of a limited self-driving capability on quiet roads in Los Altos Hills. They haven't said much since.
So the software is ready, it's just those damn regulators that are holding it back.
Come on.
The way this is phrased makes it sound like the features are just around the corner, not that the software is years from working, if it's even possible at all.
I know somebody who was told by a Tesla salesman that "Destination Based Autopilot is only months away". The advertising is absolutely deceptive, and I say this as a Tesla owner and fan of the company.
Outright lying is another matter entirely, and in some cases could even be criminal. If I have you on video telling me I can drive the car I'm buying at 90 mph toward a brick wall and the car will prevent a collision, and I then test that statement and die, the video will be used against you in court by my surviving family members.
Only where enforcement is weak. Two years ago, the government of Germany shut down Tesla's "autopilot" advertising for making false claims.
The owners who buy that are really donating a few thousand dollars to Tesla without getting anything in return. Its really strange.
A lot of people talk about the "Does it have enough sensors for Full Self Drive?" but my question would also be: Does it have the Compute?
As vehicles get closer and closer to full self drive the compute requirements have shot up, that's because a lot of the building blocks are just adding more machine learning models, pattern recognizers, and a more complete picture of the world. It is all well and good having a dozen sensors, if you can process their input.
A lot of the key cycles cannot be offloaded to The Cloud because of the tolerances (e.g. sub-second query/response for safety). One university project I happen to know of had $20K worth of compute sitting in the trunk. That will obviously come down as it is optimized but a vehicle sold in early 2018 isn't evergreen.
Where does it clearly say that? "may very widely by jurisdiction", "it is not possible to know" are anything but clear.
That second paragraph is a far worse thing. Intentionally leading people into believing that government regulators are why a technology doesn’t exist makes it sound like they’re venturing into Theranos territory.
The last sentences about selling rides sound like the sort of false concern someone would try to raise in a pitch to misdirect investors away from asking about software problems that they internally worry might be insurmountable.
http://money.cnn.com/2018/05/23/technology/elon-musk-media/i...
He liked being the media's golden boy when they were fawning over him for a decade, now things get tough because him and his company exaggerate and he wants to take his ball and go home.
(Whereas, SpaceX exiting the market would probably set back space travel considerably.)
January-March 2018: 304K EV sales, 25K Tesla sales
What the real threat is the damage to self driving that Tesla gives the public. After a series of very publicized crashes where Tesla handled their response poorly journalist are looking at every Tesla crash for involvement of their Autopilot software. Every crash is under suspicion. That alone is a hole that Tesla is going to have a hard time crawling out of and it combined with other companies missteps it will only keep setting back the field.
OTOH, that doesn't make some of his charges less accurate. The media does engage in a lot of hyperbole about Tesla crashes and always has. We don't get hyperbolic reports every time some other car has an AEB failure. We definitely don't get them about every other car's post-crash fire. And a lot of the financial opinion pieces are only marginally better.
On the other side of the coin, you have the company's misleading marketing materials. Some will say that is just as bad as the bad journalism, I guess.
> Please note also that using a self-driving Tesla for car sharing and ride hailing for friends and family is fine, but doing so for revenue purposes will only be permissible on the Tesla Network, details of which will be released next year.
Why does Tesla get to tell me what to do with my car?
If you want to drive your car for Uber, you can. If you want Tesla’s software to drive it for Uber for you, that isn’t part of the end user license agreement.
At one time automakers tried to say if you used third party oil filters or other parts, or if anyone other than the dealer did work, then your warranty was void. That didn't hold up.
Publishers tried to say that you could not re-sell books and recordings and software media after you purchased them. That didn't hold up.
Not sure that this will hold up either.
You’re going to need a copyright law overhaul to change this status quo.
For example can my water heater's manufacturer ban me from washing my dog in the bath because they own copyright on the software in the thermostat micro-controller?
If Tesla were going to provide liability insurance, I could absolutely see them having the ability to limit usage (under that insurance). But software copyright is an unusual argument.
Existing law supports this, and I don’t foresee policy changing in this regard.
If there is an operator in the vehicle, they could just enter the destination the old fashioned way.
> Existing law supports this, and I don’t foresee policy changing in this regard.
It really doesn't. I'm not sure which "law" you're even referring to.
Well, probably not, because realistically Tesla's self-driving isn't going to be happening anytime soon.
https://electrek.co/2018/05/19/tesla-releases-softwar-open-s...
The clause is probably there because Tesla wants to become the self-driving Uber, and it cannot (even if it wanted to) limit the restriction to corporations because, as we know, Uber does not have drivers or cars.
My guess:
Because they think it's their car, even after they sold it to you. And, because, more to the point, they expect that (contrary to their claims) that retaining that control and exploiting it is the only way they will be able to be profitable with the cars.
And, if you have a Tesla, you can verify this by disabling its ability to talk to Tesla. After some period of time, your car will no longer run.
Or you can crash it, get it fixed by a non-Tesla service provider, and watch it also fail to run. Not because of a mechanical failure or even a software failure, but because Tesla hasn't OKed it to run.
I'm curious about what happens if/when Tesla goes out of business. What happens to all those Model S and Model 3's out there - do they get OTA patched to no longer phone home, does the data coming out of the Tesla flip over to whoever purchases that capability from Tesla, or do they all just stop? It will be interesting to see.
What, so you'd enter your car to vacuum clean it or something, and suddenly it'll start driving to some destination it "assumed" :/
When I first complained about how autopilot was being marketed here on HN, I was down voted and told I don't know what autopilot means.
Had the exact same experience "any commercial airline pilot knows autopilot doesn't do everything for you !!!one" (which isn't even true, Cat-IIIc ILS autopilots are basically the equivalent of the car being able to do everything outside of leaving your garage and fully automated flight from takeoff to landing included was a thing in 1947: https://www.flightglobal.com/pdfarchive/view/1947/1947%20-%2...).
One could consider it a difference in what people are talking about: Tesla's tech roughly matches plane autopilot capabilities but not experience, because plane autopilot capabilities are nowhere near sufficient for unassisted autonomous driving (autopilots are assisted by automated ground-control systems amongst others).
Only a fraction of the worlds runways have Cat-IIIc capability. You could argue that a reasonable equivalent in the car world is that it can auto-drive between any two places, so long as those places are on major motorways / interstates.
Aircraft autopilots also have no way of avoiding traffic, ATC can't control them directly, and they don't have any automatic ability to do that.
On balance I'd say the abilities of Tesla's autopilot are fairly compared to an aircraft autopilot.
Unfortunately I think you are correct in that Tesla have marketed the system well beyond those abilities.
An eye for simplicity and a good set of BS detectors are not easy skills to learn. There's a whole spectrum between being a guinea-pig for the latest fad, and living like an Amish.
Having worked in the ML industry for a decade, the idea of buying anything "AI-driven' where a simple mechanical / manual option exists leaves me shuddering. Never mind autopilots and other "mission critical" systems: The first thing I did after buying our home was to rip out all the "smart home" features, and replace them by manual thermometers and switches. Year after year, I feel sorry hearing our neighbours moan about their "smart homes".
> "All you will need to do is get in and tell your car where to go," Tesla's ordering page says. "Your Tesla will figure out the optimal route, navigate urban streets (even without lane markings), manage complex intersections with traffic lights, stop signs and roundabouts, and handle densely packed freeways with cars moving at high speed."
This is very likely to be a huge balance sheet liability and probably won't be a real possibility until the cars are really old and near the end of their life. They don't give a real timetable so maybe they can get by with not delivering it for a couple decades, but it seems like that would land them in some kind of class action legal trouble.
The other day I was on 101 in the evening heading south (this is one the main highways in San Francisco Bay Area) where I was behind a Tesla on the fast lane. The Tesla in front of me was acting rather erratic. Fast jolt like corrections once in a while, etc. So I got annoyed by the erratic movements and passed the car. As I passed I glanced over. Sure enough, the driver was on his phone doing something and I'm guessing was relying on the auto pilot.
https://www.ftc.gov/news-events/press-releases/2017/05/feder...
I get that people respect and are excited by Musk's vision and work, nevermind the cars, but I feel even Tesla fans should be able to agree that the current Autopilot homepage, which is going to be the main info source for Autopilot for anyone not in possession of the owner's manual.
Someone at Tesla made the deliberate decision to have the page's headline and very first words be, "Full Self-Driving Hardware on All Cars", followed by a paragraph of text about full-self driving, followed by order buttons for Model S/Model X.
The next thing on that page is a video. It's the first and only video on the Autopilot page, and it is a 2016 self-driving demo, that Tesla has (as far as we know) not been able to push to production nor do extensive testing (assuming they test in California, which requires a mileage report).
When you buy a Tesla, aren't the purchase options for Autopilot and (future) full-self driving separate options, at a cost of $5000 and $3000, respectively? Yes, buying AP is required to buy FSD. But my point is that Autopilot exists as its own separate feature completely independent of FSD.
So to belabor the obvious, why is the only visual demo on Tesla Autopilot a 2016 alpha demo of full self driving? On Youtube I can find plenty of amateur demos of the AP features. What possible reason is there for not creating separate pages for Autopilot and FSD, other than to have consumers conflate the two systems?
Dying in a Tesla is relatively hard to do, but getting hurt or even just totaling your car because you didn't understand how AP works is bad (and dangerous for others, too!)
I feel like maybe they shouldn't turn on AP until TACC has been active for some period of time, so that you can get used to the car and how it behaves. Even now, I am sometimes surprised (like the odd behavior when stopped traffic is ahead of you a ways, and the car right in front of you changes out of your line - your car doesn't notice the stopped traffic, so you have to stop yourself or you would slam into the stopped cars). I have to assume that, at some point, the front facing camera will supplement the radar and say, "hey, that looks like stopped traffic - I should stop", but I can understand, from a software development perspective, why that's a hard problem.
I am not familiar with AP tech. So would appreciate if someone can explain why it's hard to detect stopped objects using radar.
That said, the impression I get is that Tesla's system is worse than even their initial version (provided by Mobileye).
Especially stopped traffic is a problem for vision only algorithms. It's easy to see moving objects, it's harder to see stopped objects. We humans are quite good at this, but machines have a long way to go here. Even with stereo view, everything that's a few hundred meters away doesn't have much difference between the left and right picture. Add to that that a few hundred meters away it's just a few pixels. So if it isn't moving it's like detecting a small object in a photo.
Lidar/radar are the only sensors which can reliably detect stopped cars at a distance if you ask me.
No. It's very difficult. Radar reflects off of everything. The only way to get a useful signal is to look for things that are moving at a different speed than you, like something moving relative to the road. So the radar software filters out all signals that have the same speed at the road ahead.
This means the radar won't see anything stopped on the road, because it's filtering out a huge number of returns from the road. This gets even worse because Radar reflects very differently based off the material. It goes right through plastic, water (humans) absorb it.
Even if you could try to read through the noise of the road, Radar doesn't have enough resolution to tell the difference between the road going uphill and a stopped firetruck.
"radar won't see anything stopped on the road, because it's filtering out a huge number of returns from the road"
what does that mean? sorry i am still confused.
Isn't the whole point of radar to detect objects around you?
It shouldn't matter if the object is in the road ahead, or on the side of the road, right? Because they are all obstacles that the car should avoid, no?
Like, a stopped car in front of you, vs a tree next to you on the side of the road, are both things the software should avoid, and NOT filter out, right?
It's also well known that an autopilot in a plane is no AI, it's simply following a course.
But I definitely agree it's a disingenuous name. It primarily gives the impression that you can take your hands off. So people will definitely get distracted for up to 6s which from the Uber crash is all it took. Plus beyond that you can have one hand on the steering wheel and be doing anything you want.
It's quite damming that even Uber were humble enough to realise how badly their system had failed with the fatal crash but Tesla is doubling down releasing PR statements that try to spin the statistics. I've seen interviews with Musk discussing the crash where he really seems not to be particularly bothered.
I can see Tesla putting back self driving by years if a legal clamp down comes. They are the company with by far the most miles driven with a level 3 system, but there's nothing at the moment to suggest that there won't be other crashes in Tesla autopilot cars that won't skew the statistics.
I've heard it said that self-driving cars will have to prove they're 10x safer. That takes 1B miles per death to prove. If self-driving is deemed illegal on open roads because of Tesla's recklessness it'll take much longer to prove that it is safer.
The equivalent environment of a plane's autopilot for a car is being on an empty race track. I'm sure Autopilot works well there.
Plane autopilot just automates the control inputs/outputs that are needed to follow a predefined flight plan. For anything beyond cruising or following pre-programmed flight patterns like holding loops, it needs guidance from systems on the ground that have no other purpose than telling the plane exactly where the runway is and how to descend to it. There’s not a single system involved that uses vision-based methods and/or any form of object detection. Even the most advanced plane autopilot will not be able to land the plane if the airport systems are down.
So in a way, the level of autonomy that Tesla’s autopilot provides is not that far from plane autopilot. That’s clearly not how Tesla PR sells it though.
It's most definitely an SAE2 system, not remotely SAE3. An SAE3 requires that the human driver will respond to a request to intervene in all cases where the automation system figures it's going to fail.
Tesla's is not close to guaranteeing the driver will be forewarned of an issue (at all let alone in time to intervene), the driver must be aware of their surrounding and ready to override the automation system for safe operations, the system will not guarantee safety.
With that out of the way, I am genuinely curious:
Has you or anyone you know ever believed that Tesla's have autopilot?
Anyone I have ever spoken to basically thinks that Teslas have a feature to keep you in the lane -- an advanced cruise control -- and that you must have your hands near the wheel.
Please note that even if nobody thought Teslas have autopilot, I still think deceptive marketing should be punished. I'm simply trying to gather some perspective from others, if this is actually a thought that is out there in the wild.
Since then I've lost my faith in his honesty. He's been consistently and intentionally misrepresenting what Tesla can deliver and when.
Worse, he is taking a page from Donald Trump's playbook. Faced with negative media coverage, Musk is now attacking the media and wants to build a bubble of alternative truth controlled by himself.
Rather than fix Tesla's communications, he wants to build some kind of website where "the public can rate the core truth of any article" and shame individual journalists:
https://twitter.com/elonmusk/status/999367582271422464
Can he be so naïve that he expects an Internet public to arrive at a consensus on complex issues? Of course not, he hasn't been in a coma for the past twenty years. His proposed "Muskopedia" would just end up controlled by online partisans devoted to embellishing his takes.
(Reminds me that I should probably take my $1k out. The Model 3 ship date for Europe is still a year away.)
Alright. I think it's important to be clear what Tesla is. It's certainly a very promising company, but highly successful? That it is not - at least not yet. Tesla is not profitable, and without electric car subsidies it would be in even worse condition. It has a lot of promise, but has yet to deliver on that promise IMHO.
They market it as it could be a RoboTaxi, Musk has been saying “autonomous driving is a solved problem” for 4 years now at every opportunity he gets this isn’t really ideal.
Tesla’s marketing a side the real problem is the concept of Level 3 autonomy in the first place (on the open roads). Humans can’t operate in a “hands off only” mode and it’s either all in or all out.
I’m really hoping that we could get to a good real world all conditions L4/5 (these are the same level of autonomy) before level 3 ones kill enough people to push back autonomous driving 3 decades.
Don’t get me wrong in controlled, restricted and enclosed environments for example truck yards, ports, large factories, airports, power stations, mines etc. Level 3 makes a lot of sense if you have predictable routes where trained operators know when they can be hands off and when they cannot be.
But this doesn’t and will never work on real roads with real people because eventually everyone gets “comfortable” with such a feature especially when it appears much more capable than it actually is and when Musk and many in the Tesla community make it sound like the only reason why your Tesla isn’t working for Uber whole you are at work is because of pesky regulation.