Tesla Model S waffling

[clyde]

Originally Posted by JST

Also, my guess is that Tesla will be making changes, but that it won't have final say in what those changes are. I think NHTSA is going to have a lot to say.

Tesla will make changes. More events will happen. Tesla will make more changes. Repeat many times. NHTSA will say things? In that order.

The only alternative would be NHTSA demanding Tesla turn off Autopilot until further notice. Not sure they have that authority, though. Certainly a poster case of speed of government vs technological innovation and real world change.

[rumatt]

Originally Posted by clyde

It is nearly inconceivable to me that with how smart the S is supposed to be that it could possibly lose its top and not recognize it as a catastrophic event and shut itself down immediately unless the damage prevented the car from stopping itself.

So if it loses a sensor it should slam on the brakes?

Are you sure about that?

[clyde]

Originally Posted by rumatt

So if it loses a sensor it should slam on the brakes?

Are you sure about that?

When it loses "half its brain," a number of its airbags/other systems and a significant event, yes, I'd like it to shut itself down immediately (but I'm open to arguments about something less dramatic with evidence that would be a better course).

The point, though, is that Tesla's semi-autonomous driving is not ready for use on public roads and is endangering all people that may find themselves in some kind of proximity to public roads. They clearly understand this, yet think they can shift the burden of responsible use to their customers. Disruptive business models and technologies are great. Until they start killing people in "public beta tests."

[clyde]

And the other thing about this crash, why is everyone excusing Autopilot's failure to detect the truck turning across the car's path as an acceptable failure? Because it was white and the sky was bright? The fuck? The logic so many are using to blame the driver apply equally to the car and its technology which clearly is ready to be used on public highways.

[rumatt]

Originally Posted by clyde

The point, though, is that Tesla's semi-autonomous driving is not ready for use on public roads and is endangering all people that may find themselves in some kind of proximity to public roads. They clearly understand this, yet think they can shift the burden of responsible use to their customers. Disruptive business models and technologies are great. Until they start killing people in "public beta tests."

I agree with this 100%.

At no point have I been able to comprehend how anyone thought this was a good idea.

[JST]

I agree with all of Clyde's broader points. I don't think that it makes sense to have the car slam on the brakes when it has a sensor problem, though. When the system is that compromised, adding another control input might be the right move...or it might not.

Better in my view to have it just shut down the assist and stop trying.

In fact, I think that there's a pretty good chance that the car cut the High Voltage power completely on impact. The 12V systems would have continued to function, but it's pretty likely that the main traction battery contactors opened, to isolate the main pack from the rest of the car.

You *absolutely* want the car to do that, but of course it would prevent any further motive force.

Though, as I type that, I realize it probably wouldn't affect the computer's ability to apply the brakes, which run off the 12V system.

[clyde]

If I wasn't clear, I'm not advocating having the car stomp the brakes if a single sensor fails. If multiple sensors suddenly fail/disappear, airbags fire or disappear, an impact is "sensed," unexpected and non-ordinary changes beyond a reasonable extent occur in speed, heading, yaw, pitch, and/or bank all within milliseconds of each other, the car needs to do everything it can to bring itself to a stop as quickly as possible. All has already been lost at that point. It may not be the best choice control input for all situations that could have caused those things to happen at once, but it's likely to be much better more often than anything else.

And according to Reuters, the car traveled 900 feet after impact. http://www.reuters.com/article/us-te...-idUSKCN0ZH5BW

[JST]

Should that same rule apply to all cars? And should it apply whether or not the car has Autopilot engaged?

Not saying it's not a good idea. I just don't necessarily see why it has anything to do with autopilot. Here, the problem was that the driver was killed on impact. Even if the car had been running without AP, after the crash it would have done the same thing.

In other words, whatever the proximate cause of the original impact, the car's post-crash behavior is a separate problem.

[rumatt]

Originally Posted by JST

In other words, whatever the proximate cause of the original impact, the car's post-crash behavior is a separate problem.

I'm with you, mostly. But you do have to admit there's a certain irony in the fact that it was the self-driving capability that contributed to killing the driver, then once he was dead and there was a legitimate opportunity for autopilot to make society safer, it turned itself off.

If you're going to introduce technology that sometimes makes people less safe, and you argue that's OK because it increases safety overall, there is indeed some obligation to "do good" in the broader sense. Helping bring a car to a stop when the driver is dead is a huge opportunity to demonstrate value of autopilot and help justify its existence, and the risks associated with it.

In this case it would have been difficult to ask for praise for how nicely it brought itself to a stop after killing the driver. But imagine the scenario where someone was driving and had a stroke, the car crashed into a truck and then went barreling towards a playground filled with children. Autopilot kicks in and saves the day. Suddenly the small number of accidents that may or may not have been caused by autopilot might start to seem like an OK trade-off.

Cliff notes: if Tesla is allowed to lower the bar in some cases, I'm allowed to raise it in others.

[kognito]

Wonder if anyone at Tesla has heard of the three rules of robotics

Wiki