Hacker News Reader

Comment by wccrawford | Hacker News Reader

Wow, the autopilot failed spectacularly there. This really is something that is should have caught, IMO. It doesn't even seem to have tried to avoid it, even once it got very close.

Even if you consider it a co-pilot, unless someone is paying as much attention as if they were really driving and was right in position to take control, I don't think they could have done much to mitigate this. At best, they might have jerked the car into the next lane, but even that could cause a fatal crash of that car or surrounding cars.

View on HN

Inability to reliably detect completely stationary objects is a known and documented deficiency of the system. Not making excuses, just saying it's not a big surprise.

My experience with the system is that it actually allows you to be more aware of things like this, because you can devote all of your attention to scanning the road far ahead, instead of switching between the distance and nearby details like where you are relative to the lines.

However, that requires using the system in a way that understands and accommodates its limitations, which not everybody does.

Then they should really design the system with the limitations in mind so that it can only be used like that. It seems like it allows for too much time without hands on the wheel with only a warning that can be ignored.

I agree that it would be great if the system could only be used that way, but I don't see how that could be accomplished. Having your hands on the wheel has almost no correlation with paying attention to the road. I frequently watch the road carefully without touching the wheel, and it's easy to hold the wheel while staring at the clouds or with your eyes closed.

Eye tracking? It's rather good nowadays.

> Inability to reliably detect completely stationary objects is a known and documented deficiency of the system.

Are you for real? Subaru's Eyesight would have detected this, and it is much less sophisticated than Autopilot (just two visible light cameras).

In the first crash Tesla explained how it occurred (strong sunlight blocked the visible light camera, and radar failed since the truck appeared to "hover" over the road) but they haven't explain this crash. Why didn't the visible cameras cause auto-breaking?

Breaking for stationary objects is the PRIMARY purpose of auto-braking systems.

Yes, I'm for real. Direct quote from the manual:

"Warning: Traffic-Aware Cruise Control can not detect all objects and may not detect a stationary vehicle or other object in the lane of travel."

> "Inability to reliably detect completely stationary objects is a known and documented deficiency of the system."

"Caution: car may drive into solid objects at 70mph"

Meanwhile on the front page of HN, a system which is not calling itself autopilot manages to avoid flying a plane into the ground: https://news.ycombinator.com/item?id=12501985

Technologies like self-driving/autonomous/autopilot cars need to be very clear about what they can and cannot do. If the human has to be paying attention and able to take over before a solid object appears within the stopping distance of the car, then it should not be marketed as any of those and should limit itself to calling itself a driver assistance device.

I would just like to point out that your caution applies to just about all cars currently on the roads.

Yes, but they're not marketed as being self-driving.

(This whole issue is very important for people who do want fully autonomous cars, as Tesla et al messing up the PR will get them banned if they're not careful.)

This car isn't marketed as self-driving either. Some people are inferring that from the name, but Tesla is fairly clear on the system's actual capabilities.

Incidentally, Mercedes does advertise some of their cars as "self-driving":

https://electrek.files.wordpress.com/2016/07/mercedes-self-d...

I wonder why we don't hear these complaints about them.

You didn't hear these complaints. They where made, and Mercedes pulled the ads. http://www.autoblog.com/2016/07/29/mercedes-pulls-self-drivi...

There's no evidence that Autopilot was enabled. It won't perform any avoiding action if it is not enabled.

There is no evidence that Autopilot was disabled either.

No. But it is misleading for media to report "autopilot" caused a crash when we don't know yet.

Which media did ?

They did the same thing when a woman plowed into the side of the building at 30MPH. Obviously Tesla said it wasn't enabled, but reporters didn't wait to hear back from Tesla.

Reply on news.ycombinator.com