Waymo – Avoiding a Falling Skateboarder
twitter.comImpressive feat by Waymo, though it could have been 100-200ms faster had it noticed not only the change in trajectory, but also the jerky motion that destabilized the scooter.
For a truly superhuman performance, it might even register the brick and anticipate the scooter's accident.
But what would the car choose if there was no room to the left? Would it choose to hit a car next to it, or hit the person?
I'm curious how no-win scenarios are programmed into it and decided.
I'll admit I have not done any research into this and I am sure I could.
Hopefully it generally avoids no-win scenarios. It could be that it decided it was safe to drive that close to the scooter because it had an avoidance option. If there was no avoidance option it should stay further from the scooter to maintain other options, like coming to a complete stop before hitting the scooter.
> Hopefully it generally avoids no-win scenarios.
But unavoidable no-win scenarios exist on the road. "What does the vehicle do?" is a reasonable question to ask.
In the case of bike lane traffic or heavy pedestrians, if no avoidance options are available, your suggestion could easily restrict traffic to sub 10 mph. (E.g., to be able to stop with minimum impact if a pedestrian suddenly darted into the street.)
(I am glad to see that in this video, the Waymo crosses a double-yellow, using a dead zone between the two sides of the road to avoid the collision, thus bending the rules to avoid the hit, which is great.)
There are gradations. Yes, it might be unreasonable to never get in a situation where hitting a pedestrian is impossible, but it's not unreasonable to expect to never get in a situation where hitting a pedestrian at more than 20mph is impossible. Below 20mph a pedestrian collision is usually non-fatal.
That might mean limiting speed to 30mph or lower in residential area. That is fine.
Just like airplanes and surgery and almost every other dangerous activity, it should be impossible to have a road fatality with only a single failure. Only multi-systems failure should result in a fatality from any form of technology.
> I'm curious how no-win scenarios are programmed into it and decided.
I hope the pedestrians always have the highest priority, but then what if there are two of them... Soon they'll use bluetooth to pull your net worth from your apple watch and steer towards the smallest number.
I hope not re: pedestrians.
An adult running onto the highway is very different than a young kid doing the same.
All else equal, I would not preserve the adult's life over that of drivers. I would preserve the kid's life over that of drivers.
Why? The adult is presumed to be responsible for the repercussions of its actions. The drivers did not create the problem. The family of the drivers could hold others accountable for any unreasonable systemic danger leading to the death of the drivers.
wow. Liabilities
The Trolley Problem! Kill the skateboarder who fell or the driver in adjacent car. Establish a situation where hard breaking is not safely possible. Make the AI choose. Not jokingly, the choices in such scenarios should be disclosed.
I think those scooters top out at 15 mph, so hard brake might have been an option.
There might be a time when it's socially acceptable for self-driving vehicles to have [air bag-like?] cattle pusher fronts to mitigate unavoidable contacts.
If there's a car to the left, the choice is probably between hitting only the person, or hitting both the person and the car to the left.
I wouldn't expect you'ld be able to push the other car over enough to avoid the scooter rider.
In that situation, full brake application is probably all you can do.
You’re only thinking in the first order.
In reality no-win scenarios are avoided by having so many cameras and such an overwhelming amount of data that Waymo would win in court instead.
This is good for all as the incentive for maximum data collection as CYA also means maximum velocity of product improvement.
Cutting edge technology solves a problem that shouldn't even exist in the first place episode 376: bicycle infrastructure edition
PS: it's an rental e-scooter, not a skateboard, another product of our piss poor infrastructure