[mymacbook]

Reading that initial report is terrifying. I am so glad the NTSB set the record straight that the driver had his hands on the wheel for the majority of the final minute of travel. Really makes me feel like Tesla was out to blame the driver from the get go. To be clear the driver is absolutely partially at fault, but my goodness autopilot sped up into the barrier in the final seconds — totally unexpected when the car has automatic emergency breaking.

Emergency breaking feels not ready for prime time. I hope there are improvements there. Don’t want to see autopilot disabled as a result of this, would rather Tesla use this to double down and apply new learnings.

Just so sad to hear about this guys death on his way to work - not the way I want to go. :(

[kevincox]

> Don’t want to see autopilot disabled as a result of this

I would. I think any self-driving system that expects the "driver" (passenger I guess) to take over in tough situations is bound to fail again and again. If people feel like they don't need to pay attention they won't no matter how much they know that they should.

I think that partial autonomy is a sort of uncanny valley, either side is much safer.

[prawn]

I'd prevent them from using the AutoPilot term, penalise them for marketing imagery that includes people driving with their hands off the wheel, and require hands to be on the wheel far more often.

Until it's significantly improved, it should be a backup safety system rather than a driving and safety system.

[Robotbeat]

>I'd prevent them from using the AutoPilot term

I disagree. It's the same term as in aviation and does the same thing.

> penalise them for marketing imagery that includes people driving with their hands off the wheel

That's more reasonable.

[HumanHater]

Can this "Autopilot" control altitude of the vehicle? No. Therefor it's not the same thing, it is just something similar.

What should we call car autopilot? There is more than one opinion. Some people believe in importance of some weird technical characteristics. For most people defining property of autopilot is its ability to control vehicle while pilot left the cockpit for a few minutes. So to them car autopilot must safely drive some roads and require human to take full control on the others. Taking over in case of emergency doesn't exactly fit this idea.

What looks really weird in all this discussions is their fanatical uselessness. Who would be hurt if new law would prohibit to call autopilot anything less than L5? Well probably no one. Who would benefit? Most likely a lot of people. Why would anyone oppose regulation like this? Huge question.

[hashbig]

> It's the same term as in aviation and does the same thing.

Except in aviation trained pilots with at least a decade of experience flying planes get to use it, understanding well the limitations and purpose of the technology. A teenager can use Tesla's "Autopilot".

[Robotbeat]

Incorrect. Autopilots are commonly used even in general aviation, and you can get your pilot’s license in just a month or two if you had a cooperative flight school. Heck, there are accelerated immersive programs that train you in just 14 days.

You only need ~40 hours (minimum) of flight time to get your private pilot’s license. This isn’t even much different than driving license requirements, except that planes are a lot more expensive to rent and operate.

(And for the record, teens can also use autopilots on airplanes since you can get a student pilot’s license allowing you to operate a plane by yourself at age 16. Roughly the same number of hours of supervised instructional operation as getting a driver’s license.)

[rurounijones]

I think you are nitpicking definitions here when it is not the issue.

The issue is that the general public think of "Autopilot" as the magical piece of technology that automatically pilots thing while you sip a pina colada and otherwise switch off mentally.

What the 0.5% of people who actually know what is required of the flesh and blood pilot when using an aviation autopilot is not the issue. It is what the majority understand.

Changing what Tesla is allowed to call it is the path of least resistance rather than trying to educate the general public on "what the team actually means in the industry that made the term famous and therefore should be universally applied elsewhere".

[prawn]

As I understand it, aviation autopilot is generally used in open air where there are typically not frequent stationary objects the software is not designed to detect. Also usually sold in a different way to a different market.

[testplzignore]

> Don’t want to see autopilot disabled as a result of this, would rather Tesla use this to double down and apply new learnings.

I agree that Tesla obviously needs to be learning from this. But at this point in time, Autopilot needs to be turned off until Tesla can fix the fundamental flaws in the set of sensors on their cars. There is now strong evidence that Autopilot is dangerous, and there is still weak evidence that it is less dangerous than the alternative (i.e., a human driving).

I expect that the NTSB will eventually issue a safety recommendation that Autopilot not be used in its current form. I'm a bit surprised they didn't issue an urgent recommendation in the preliminary report, or even before.

[TaylorAlexander]

I have a hard time seeing it this way. Turned off until they can fix the fundamental flaws? Which flaws exactly? It’s a prototype technology that no one knows how to do perfectly. Should Waymo also be taken off the road? None of these technologies are provably safe, so we always have to accept some risk. How much risk is appropriate is hard to determine. Perhaps LIDAR would reduce the risk, but Uber showed you can kill people even on a LIDAR based vehicle if your software fails to classify the risk properly. And humans drive with just two eyeballs so it’s not unreasonable to think camera based systems should be in principle possible to make safe.

So we could pull all the self driving cars off the road, but then we inhibit their ability to collect the wide range of real world data necessary to build the technology. Or we could pull the less successful tech off the road sooner, but that would cause those programs to wither and hand the market to one or two well funded players.

I just don’t see how pulling the tech off the road because of a flaw that Tesla has likely already fixed would accomplish much.

I do struggle to understand what we should do, but I’m not so quick to think the tech should be pulled.

[falcolas]

> To be clear the driver is absolutely partially at fault

I'm... not so certain. Why? The autopilot had likely exhibited proper behavior every time that the vehicle had passed that particular section of road prior, and if the driver was paying full attention to the behavior of the vehicle, he would only notice the problem around the 5 second mark.

Five seconds, if you have no reason to be concerned about the vehicle's behavior, is not much time - especially if you consider that alert drivers are recommended to give themselves a minimum of 4 seconds of reaction time (i.e. follow a vehicle by at least 4 seconds).

[ericpauley]

My vehicle (a Honda civic) exhibits this exact same functionality and behavior. (lane keeping, ACC, emergency braking for cars) They make it very clear the limitations of these systems. I'd say that in the 5 months I've owned it it's had this exact behavior (veering off exit ramps) 10 times. A simple jerk of the wheel puts it back on track, it's such a natural motion if you're paying even the slightest bit of attention. That being said Tesla fails to make their drivers aware of the limitations of autopilot, so I agree that this may not be in the driver.

[freerobby]

Tesla reminds you to pay attention and keep your hands on the wheel every time you engage Autopilot. It's one of very few legal disclaimers they show you all the time on the screen, and don't give you any way of turning it off.

[jonhendry18]

And we all know how assiduously people pay attention to messages that flash up on screen.

[freerobby]

How about audible nags? How about flashing white lights on the display? How about gradually slowing the car down until you give tactical feedback proving you are in control? Tesla does a lot of things to coerce drivers to pay attention. If you check out TMC, you'll see lots of people complaining about how paternalistic and "naggy" the system is, even for those who use it properly.

I am continually surprised by how little emphasis there is on personal responsibility when this community discusses an L2 system such as Autopilot. According to both the law and the operating manual, the driver is in control at all times. Tesla warns you of this every time you turn it on. Yes, there are enough bad drivers out there that Tesla is wise to implement habit-forming nags; but drivers also need to take responsibility for how they use (and abuse) these systems. Nobody would pass the buck to cruise control for a driver who set it to 65 and then plowed into something in a moment of distraction. All due respect to the victim here -- and I feel absolutely terrible for him and his family -- but if you are paying attention and looking at the road ahead of you, there is no situation where you accelerate for three full seconds into a concrete barrier at 70MPH -- not with Autopilot, and not without it.

[dralley]

If I recall correctly (and I might be wrong), didn't the driver in question actually report prior to the accident that autopilot had anomalous behavior on this section of road?

[freerobby]

> I'm... not so certain. Why?

In this specific case, because if a driver is watching the road in front of them, four seconds is an awful lot of time to watch yourself head toward and then accelerate into a cement barrier, all without touching the brakes. It's fine to want Autopilot to be better, but as a matter of law, it is the driver's responsibility to slam on the brakes in that situation, and move into a legal lane.

More broadly, it's because the contract of an L2 system is that a human is in control at all times. L2 systems are assistive and not autonomous. They will never disobey a driver's presets, nor override a driver's real-time inputs, even if the car thinks it is safe(r) to do so. This is a major design principle behind every L2 system, and the reason why there are no scenarios where an L2 system is considered at fault by law.

Now obviously, if there were a bug that caused an L2 system to override a user's input -- say disregarding someone hitting the brakes, or overpowering the steering wheel, that'd be a systemic failure and grounds for a recall. But we haven't seen any cases of that, and there are hardware precautions (e.g. limited torque in the Autosteer servo) to minimize that possibility.

[jonhendry18]

"The autopilot had likely exhibited proper behavior every time that the vehicle had passed that particular section of road prior"

The driver had actually reported to Tesla problems about that specific section of road.