Hacker News Reader

Comment by exotree | Hacker News Reader

This is a very upsetting response. Yes, the driver could have paid more attention, and yes, he shouldn't have activated Autopilot in an area where he knew Tesla's autopilot was not working correctly.

However, Tesla calling it "Autopilot" should be no longer allowed. It's clear that it is most certainly not that, and customers are not treating this feature in a careful and cautious fashion, and I do believe that's due to Tesla overselling Autopilot's abilities. In addition, its warning systems are obviously not enough to get drivers to take corrective action and ensure their hands are on the wheel. (Why you can take your hands off the wheel at all when engaging autopilot is a mystery to me.)

View on HN

It'll work 99% of the time and lull you into a false sense of security. The other 1% of the time you better be ready to take over with little notice or there's a chance you'll end up dead. And with automatic updates a section of road that has been fine for a year suddenly tries to kill you...

Blaming the victim certainly doesn't sit well with me.

I totally agree. I am a massive fan of Tesla but this response is absolutely atrocious and disrespectful.

They refuse to acknowledge a bug in THEIR software which drove his car in to a concrete barrier and pin all the blame on him. Even so far as to tease him and ask why he would even use their product in that area (which Tesla allows).

It must feel so so so sad to have your loved one die and then to read this heartless BS from tesla.

I also always say this to others. It doesn't matter if it works 99.9% of the time. It took them maybe 10 years to get the software to this point. But it will take 50 years to solve the rest. And nobody is going to use the systems before that.

Autopilot is a marketing term here. It's correct if we think of Autopilot = autonomous but it isn't. The thing alarms at you (visually and via sound) when you take your hands off the wheel. The fact that drivers ignore that isn't Tesla's fault, and saying it's "not enough" is kind of absurd. What the hell are they going to do if you take your hands off the wheel? Stop the car in the middle of a freeway?

I could use your exact same argument to say Ford isn't doing enough to force people to put on their seat belt. It's a silly argument. It provides seatbelts, alarms when you fail to put it on. At that point, any harm is user-error. That's the exact same case here.

> Stop the car in the middle of a freeway?

Yes. When the alternative is having a crash they should find a way to stop. [edit: of course I don't mean to suddenly stop. But if the system can't trust the driver is alert it should disengage or stop. As we've seen now multiple times, if it continues it will result in a crash.]

The current behavior is not only dangerous to the driver of a tesla. It puts everyone in the road at risk, most of whom have no contract with tesla and never agreed to be guinea-pigs in their so-called "autonomous" experiment.

Stopping in the middle of the road is dangerous (people will crash into you) and it would also be hard to make something have a feature to pull them off the road. If the computer reads it wrong you'll be in a ditch from the car trying to pull you off the road into something with no shoulder. The shoulder of a road is also dangerous as well even if there is a proper one because people often do not stay exactly in their lane. This is why officers often come around to the passenger side.

Just make the car turn on the warning lights and gently come to a stop when the driver lets go of the steering wheel or is otherwise not paying attention.

There are two possibilities:

1) The driver is actually incapacitated. Then slowing down and stopping is a million times safer than continuing to drive.

2) The driver was looking at his mobile phone. Then he'll probably quickly react and take control of the car again.

I don't see the scenario where continuing to drive is safe when the car knows the driver isn't paying attention.

Big city highway traffic where the road is crowded and moving fast. Those people aren't paying attention either and will slam into you. I've seen cars stopped on the freeway a few times in Atlanta because they broke down and it causes a big wreck. Responders really rush to get them off the road because it's so dangerous.

The biggest issue here is 90% autonomous has been shown to not work. People invariably let down their guard and go "hey, this works" and promptly stop paying attention because there's nothing for them to do.

>People invariably let down their guard and go "hey, this works" and promptly stop paying attention because there's nothing for them to do.

This is not specific to Tesla.

Trivial example: A coworker has a car that has blind spot detection. She knows the manufacturer says "you should still check, it is not 100% accurate". And she admitted that despite knowing this, she does not check her blind spots and relies solely on blind spot detection.

More serious example:

I have a car that has adaptive cruise control. You can get on the highway, set the CC to 60 mph, but if traffic slows down, it automatically slows down and maintains a distance to the next car. It speeds up to 60 mph when the traffic picks up again.

I love it, but for the inexperienced driver, this is dangerous. It requires one to be more vigilant than regular CC. Every so often (rare, but has happened a few times), it will not detect that the vehicle in front of me is a vehicle, and will not slow down. The more you rely on it, the more vigilant you need to be. My wife has never used CC before, and this would be too much for her to handle, compared to simpler CC.

It's a general problem. Tesla is just a convenient target.

I very much agree with your opinion. Systems which take over most but not all responsibility seem to lull us humans into a false sense of security/reliability. The psychological aspect of this is actually quite intriguing, but it's also a very dangerous phenomena.

I don't know anything about your wife, but that does sound perhaps a touch condescending. Besides that I think you make a very good point, one that many people seem to want to wish away.

>I don't know anything about your wife, but that does sound perhaps a touch condescending.

That's probably because you don't know anything about my wife :-)

This isn't about my wife. I would never recommend anyone to learn CC on this type of system. They should learn CC on a regular car, and once they're comfortable with that, switch to something like this. You simply have to be more alert for a "safe" CC than a regular one.

> alarms when you take your hands off the wheel.

No, only every ~ 1 minute.

>What the hell are they going to do if you take your hands off the wheel

Detect barrier and apply auto emergency brakes and Disengage autopilot like GM's Super Cruise does if it detects your face is not looking at the road.

>However, Tesla calling it "Autopilot" should be no longer allowed.

Meh, autopilots on boats don't do anything more than steer a straight compass course.

That's inaccurate.

Many marine autopilots integrate with AIS, sonar, and charts for accident avoidance and alerts. For example on some larger ships it is able to calculate if at the current speed a collision would occur and reduce speed to avoid it. Some can do "port to port" auto-steer, taking the vesicle through narrow channels.

We are both accurate. I was talking about boats, not ships. My point still stands - ships are required by Collision Regulations to keep a lookout at all times. And yet they are called "autopilots". Even the most crude of devices that I described (that is used on pleasure boats).

A big part of being allowed to take your hands off the wheel is probably "how do I detect the hands are on the wheel?". In current S/X with no interior camera the way autopilot works is you have to "yank" the wheel every now and then to prove you have your hands on. Requiring constant "yanking" wouldn't work, and there are often long stretches of roads on highways (where Autopilot is meant to work) that wouldn't require any movement of the wheel at all so they wouldn't know the difference between hands on or off unless you "yanked" it.

Every time I learn about how these cars are handling things I end up disappointed and concerned because the actual implementation is way less impressive and seemingly less foolproof than I - and I'm sure many other people - are assuming they are.

I would have never guessed "hands on wheel" is detected only if the driver yanks the wheel. I figured it was camera based or that the wheel had capacitive sensors.

The steering wheel detects torque. Generally if your hands are on the wheel, they exert enough torque through natural movements for the vehicle to detect them. In rare cases, where a road is straight and a driver's hands are perfectly balanced, the car will flash a warning, in effect telling the driver to provide some brief feedback. The slightest of forces on the wheel will quell the warning (much less force than is required to move the steering wheel).

Are you talking about vehicles in general or Teslas? I've had a Model X for about a year now and can say that just having hands on the wheel definitely does not generate enough torque to deter the warnings on Autopilot, it definitely requires a "yank".

Teslas. Maybe something changed but on my 2015 Model S I've never had to physically move the wheel in the slightest, just apply a little bit of pressure.

Now compare this with Tesla's PR statement which makes it sound as though they have proof that the guy who was fatally steered into a barrier by their Autopilot had taken his hands off the wheel. They resort to this kind of sleazy, dishonest misrepresentation every time their software screws up.

Cadillac Super Cruise is far superior to Tesla's autopilot and is a true Level 3 system. It does active attention monitoring of the kind you're expecting and uses GPS fences to only activate where GM has detailed maps.

Still very new only only on a single model so far, but it's at least the right direction.

https://www.youtube.com/watch?v=VTuiB6Ob7Jc

I also don't hold the wheel hard enough, I always get the jiggle the wheel warning ~ 1 min as that's the only way for them to detect. Sometimes if you're so focused on watching cars in other lanes it's easy to miss the slight flashing warning that it starts with. So there is no evidence Walter wasn't paying attention in the past just based on that as most AP users get that.

I see your point on calling it "Autopilot", but if you think about the application of the term in terms of aviation, I think it makes more sense. Autopilot in planes won't land the plane for you. It basically will just keep you at a certain elevation and direction.

I think that Tesla's Autopilot function is comparable to planes. That's probably how they will justify it. However there might be a legitimate point in that people don't have that kind of understanding of what "Autopilot" means.

Aviation autopilots handle flying duties in a giant expanse of open sky. It's a way, way easier job to do than navigating a highway from behind a steering wheel.

So even if the "capabilities" may be comparable between the two systems, the amount of supervision they require is not, so I don't think they should use the term. It may be "lane-keeping cruise control" isn't nearly as sexy, but it doesn't establish unrealistic expectations which get people killed.

> Autopilot in planes won't land the plane for you.

For unmanned aircraft, autoland is an safety feature. After loss of link, an autopilot might be configured to land at a designated location, rather than just circling until it runs out of fuel and drops out of the sky. Though admittedly, crashing in a safe place is less serious when the vehicle has no people in it.

The Tesla autopilot needs much closer supervision than any of the autopilots I've worked with. That's more a result of its domain than of its sophistication, but it still makes me very uncomfortable with the name.

Autoland is also a safety feature for manned aircraft because it allows you to land in bad visibility or avoid accidents like the one at SFO a couple of years ago.

Excessive reliance on features like autoland was actually the cause of that SFO accident, according to the NTSB.

But ILS (the autoland system on ground) was disabled that day. Are we talking about two different incidents?

No, did you read the accident report which talked about the pilots being unable to land the plane because they were used to the computer doing most of the work?

Well, people in aviation are more aware (hopefully) of the limitations of the systems and even then gets on snags.

I agree, Tesla shouldn't call it autopilot, call it Highway Assistance or something.

Or even, at this point, I don't see why it should even exist, if it fails on basic situations.

Actually autopilots in airliners, and some military jets ARE capable of landing aircraft.

And have been doing commercially so since 1966!

First Cat III autoland aircraft was the Shorts Belfast freighter. Sud Caravelle and HS Trident airliners started at lower categories and worked up to III.

In a domain like marketing, it doesn't matter what something technically means, but what people think it means.

Autopilots can and do land planes; the first autopilot landing was decades ago! They simply don't as a matter of practice so that pilots can maintain their skills during landing, which is the most crucial and dangerous part of the flight.

Additionally...Autopilot in planes is hands off. The pilots look at the gauges but they're not actively maintaining their grip on the controls.

Now you're damning Tesla because of a safety feature: the point of keeping the hands on the wheel is that it's a way of keeping the driver alert.

No, I'm damning Tesla because they sell the vehicle as self-driving when it cannot, and moreover "require" the driver to maintain superficial alertness without actually enforcing alertness in a reasonable matter like all of their major competitors do.

Wow. You sure are an expert in alertness.

Reply on news.ycombinator.com