Hacker News Reader

Comment by arthuredelstein | Hacker News Reader

I wonder how this will work with self-driving cars. Making eye contact is not an option.

The self-driving car detects a human at the edge of the intersection and hesitates briefly. If the human begins crossing, it waits for them to cross; if the human doesn't, the car proceeds cautiously through the intersection, ready to brake smoothly if the human suddenly steps out.

Yup, and furthur refinements can detect intent, hesitation, emotion, and other kinds of social signalling. (Drivers, bikers, and pedestrians will use non-verbal cues to negotiate; this happens in other contexts too)

The first court cases where someone purposefully throws themselves in front of a moving self driving car will be interesting. Obviously I hope this never happens, but it's inevitably going to happen at some point. Hopefully the sensors are gong to be good enough to be able to save those individuals with no injuries. Though I guess from the car's perspective, this is identical to an animal or child darting into traffic, so this will have to be accounted for in some capacity anyway.

Is there training data for suicidal pedestrians? Does it even happen often enough for a vision network to pick up on those queues?

People purposefully jumping in front of (slowly) moving cars is enough of a problem in some parts of the world that basically every car has a dashcam now. Not suicidally though but trying to get compensation.

Well, the sensor system in a self-driving car conveniently fills the role of a dashcam as well. Logged sensor data will make many accident investigations much easier.

What if as the owner of the car I don't want the sensor data released? Will self driving vehicles move toward a (much of) software like model where we only are getting a temporary license and some other entity can change the rules out from underneath? I'm really excited for self driving cars, this is just one more of the many interesting legal/social questions I look forward to seeing debated.

It is an option, and I think a pretty effective one in the future. If advanced enough, self-driving cars could learn the different cues we give each other to communicate (be it with a car driver, a bike rider, a pedestrian, or any other agent).

It will be very interesting to see how this dynamic evolves over time.

For it to really be equivalent to eye contact, the car needs to somehow signal that it is aware of the pedestrian and their intent.

I can only hope they'll show old-timey clips on a screen, like Flubber's Weebo.

Self-driving cars won't need the glare, because they will be programmed to respect right of way, including but not limited to pedestrian right of way, or the manufacturers will be crushed by recall/repair/replacements costs and defective product liability lawsuits.

Individual drivers are a bigger enforcement problem.

I've watched self-driving cars trying to make a right turn get "stuck" behind a continual stream of pedestrians. Human drivers will eventually aggressively poke through, but self-driving cars are a little too timid for that.

They could move forward very very slowly

That strategy sometimes works with crowds (e.g., home game weekends in college towns). But sometimes, things turn ugly.

Glare at its forward camera.

That can work if there is also an anthromorphasization (put a human-like face that people can relate too). There was another article talkinf along these lines -- a disabled person trying a telepresence bot for the first time at a tech con saya he did not get the usual courtesy or respect. I suspect people didn't pattern match on it being human.

(I also found "glare at its forward camera" funny and ironic -- we humans have a lot to learn about relating to non-humans.)

Reply on news.ycombinator.com