Hacker News Reader

Comment by netinstructions | Hacker News Reader

Tesla's statement[1] provides some more details

> What we know is that the vehicle was on a divided highway with Autopilot engaged when a tractor trailer drove across the highway perpendicular to the Model S. Neither Autopilot nor the driver noticed the white side of the tractor trailer against a brightly lit sky, so the brake was not applied.

Failing to detect the white side of a tractor against a brightly lit sky is something I can see camera/image based sensors struggling with, but not LIDAR based sensors.

[1] https://www.teslamotors.com/blog/tragic-loss

View on HN

There was some discussion, but Tesla has a forward facing radar, but it is mounted lower than the camera and trailers of certain height have been shown to avoid it's detection window.

So, speculatively, it is a double failure:

1) Trailer was high enough to avoid radar

2) Trailer was light enough/low contrast to avoid detection by the camera

Upside of working on "boring CRUD apps that won't change the world": your edge-cases won't result in fatalities.

I hope Tesla learns something from this, from the outside looking in - they appear to have a cavalier, can-do attitude (nothing wrong with the latter) towards safety engineering. Release beta self-driving software to production vehicles? Yeah, why not?

This is going to be a major morality issue with self driving cars.

What if a deer crosses the road? Will the car hit the deer to save the life of the driver? Avoid the deer to avoid an accident? Or save the deer and run into the ditch, possibly harming the driver?

Unfortunately, this technology is not capable of making those decisions and if ultimately the driver is responsible, then maybe we should just keep it that way then.

"Or save the deer and run into the ditch, possibly harming the driver?"

This is not even a slight morality issue. The person is more important than the deer.

"Unfortunately, this technology is not capable of making those decisions"

Of course it is. Program the machine so that its primary concern is protection of human life. Which should be its secondary concern (protection of property or protection of deer) is a more interesting question. But running into the ditch to save the deer while threatening harm to the driver is simply not an option.

Replace 'deer' with 'child' and there's suddenly more of a quandary

Rapidly apply brakes. Seems like the safest possible way to avoid the deer without causing injury to driver + possibly others.

That causes the front end of the car to pitch forwards, which can actually increase the severity of a collision with large animals.

Doesn't it also cause the car to stop?

I think the problem might have been that the middle of the trailer is not at ground level but at some hight where the lidar data is ignored or not there.

How does Tesla know the driver didn't see the trailer? He was killed in the crash. Is there a cockpit voice recorder?

Presumably because the brakes weren't engaged.

It is possible he was expecting the Model S to auto-brake as it normally does, and possibly realized when it was too late and panicked/tried to swerve. But I'm just speculating, I'm waiting to hear the outcome of the NHTSA investigation.

I'd have thought the auto-brake would be expected to react faster than a human driver in that situation. So if you saw such an event and the car wasn't already braking you'd hit the brakes, not sit around waiting to find out if it handled it or not.

Reply on news.ycombinator.com