Hacker News Reader

Comment by Smaug123 | Hacker News Reader

In fairness, if you put enough fake markings on the road, any human will get confused too. The important thing here is how small/few were the fake markings, not the fact that this was possible at all.

View on HN

Of course if you put bad input into a system you are going to get bad results, whether human or computer. You periodically read about dumb kids messing with stop signs and causing accidents.

https://www.abc12.com/content/news/Stolen-stop-sign-leads-to...

https://www.kansas.com/news/nation-world/national/article218...

https://www.foxnews.com/us/charges-dropped-against-1-in-stop...

https://www.washingtonpost.com/archive/politics/1997/06/21/3...

Humans have enough intelligence, though, to be able to figure out when the other lane is oncoming traffic and those spurious markings are in fact spurious (snow or a plastic bag).

Sure, but a self driving car might know it should stop at an intersection even if the stop sign is missing because it is a database of stops. Computer and humans are going to have different weaknesses, but if you are intentionally trying to fool them you are going to be able to find a way to do it.

Yes, and so we learn to never trust user input when writing software because it might be malicious.

But not trusting sensor input for cars is not really an option, it's all you have, and everybody can try to change it.

True, but it is no different for a human driver. All we have is our senses, and anyone can try to fool them.

The thing that concerns me more is the machine's confidence. A human seeing unusual lane markings would be confused and exercise extra caution. Confusing a person is easy, but I don't know how you could make them confident there is a nonexistent lane.

I think you just explained to me why I've found robots/machines to be dangerous when they are operating. And that's because they don't have any caution. They will accelerate swiftly, operate at high speeds, and all that. They do it all well, until something goes wrong. Suddenly, having no caution becomes a huge liability.

Not where it snows. I grew up in a place with real winters and agressive plowing/salting of roads. Drivers there don't rely on what's painted on the roads since it's often obscured and the state DOT doesn't really bother putting much information on the road surface, since it's often obscured and when not obscured too damaged from the plowing and salt to be of reliable use anyways.

It was one of the notable differences when I relocated to CA. CA puts a lot more information on the road surface, and CA drivers correspondingly rely on it more heavily.

Not really. Traffic on 95 in NC/SC/GA keeps moving at 50 even in southern downpours so heavy the lane markings are invisible. Here in DC the lane markings are ordinarily entirely off, because lanes get closed when there is construction on a building by the road. People figure it out. Self driving tech that relies on lane markings isn’t viable in the real world.

Humans have millions of years of evolution, particularly at selecting for predictive behavior within real world physics (sight the prey, lead the shot, kill the prey; avoid the predator, etc).

Combine that with past experience about what the shape of the roadway is, and a human can fill in a LOT of missing data and correctly read clues to get a 'good enough' update on the situation and move forward.

Usually that's along the lines of: follow the car in front of me, and match their changes in motion to avoid the badness.

As long as things work, which they will most of the the time on a freeway where no one's screwing around, you'll get out of the other side without issue.

There is a road on my way to work that has a darker asphalt line that goes diagonally across the road for about a quarter of a mile. Inevitably, I and others follow this line before realizing we are leaving our lane.

Especially when the sun is low on the horizon and the glare off the road surface makes it hard to distinguish markings from changes in surface sheen.

I never drive without polarized lenses now.

Yes but humans are orders of magnitude more adept at adapting to scenarios they have never seen before

Humans have a much broader suite of capabilities, usually allowing us to a) recognize that this is happening and b) work around it.

A human can be confused for sure, but humans learned to see in a non-standard situation what the drivers ahead are doing. If all drivers ahead keep going straight even when road markings are confusing and may look like a lane change then most drivers will also go ahead.

AFAIK no self-driving technology can do this.

Reply on news.ycombinator.com