[frobozz]

> Does it really try to strap the person to a hospital bed, underwater...?

It looks like it folds two barriers against the sides of casualty to pinch them in place. I imagine that if you are not completely incapacitated, you can dodge or wriggle out of it. I would also guess that, given that it's supposed to be clever enough to spot an incapacitated casualty to rescue, it might be clever enough to stop trying to rescue you if you attempt to swim away.

> Won't that drown the person if the robot is too slow or malfunctions?

WRT rescuing unconscious casualties from dangerous situations, you start with the assumption that they will die if you do nothing, and could already be dead. With that in mind, who cares if this robot drowns someone halfway to the surface who would otherwise have drowned at the bottom?

> So the reasoning is: job training for lifeguards is not feasible (why?), so we should develop super expensive robots that local pools can't afford.

I agree with you on this point. I think there is something in having a system that can help spot swimmers in danger, but then leave the job of rescuing to actual lifeguards.

The problem of not having enough lifeguards is not that they cannot perform rescues, but that they cannot continuously observe the entire area and spot people in need of help. Apart from spinal injuries, a single lifeguard is perfectly capable of performing a rescue in most pools or swimming ponds (in much larger or more complex areas, like water parks or the sea, it might be harder to get to a distant casualty from any given station).

Another thing is that lifeguards prevent the situation that this robot is designed to rescue people from. You spot a swimmer in trouble, and toss them a float or a line. You spot a dangerous situation arising, and you take action to prevent it. You don't wait for them to sink to the bottom and pass out.

[craftinator]

> I would also guess that, given that it's supposed to be clever enough to spot an incapacitated casualty to rescue, it might be clever enough to stop trying to rescue you if you attempt to swim away.

One would hope, but I wouldn't hazard a guess at that. It likely uses some type of ML/NN to spot an incapacitated person; there are a bazillion pretrained models that recognized people in images. I know of no models that recognize when someone is "trying to escape". I think that that would be quite a difficult problem to solve, except for possibly a number of panic buttons. But in general, guessing a person's intent programmatically is really hard.

[frobozz]

For decades, all sorts of devices with very simple processors have been able to detect when the force against an actuator exceeds a certain amount and reverses or disengages it.

[craftinator]

Yes, but how do you infer intent from the force put against the actuator? There is no point in using this robot if it doesn't work if people are heavier than some arbitrary value, or if they get released because they bumped against something while being towed...

[frobozz]

Also, if it can spot an incapacitated person amongst all the other swimmers, then it must be able to recognise when people are not incapacitated, otherwise it would just be going around randomly grabbing swimmers.

[klmadfejno]

I give it 7 seconds before the children start pretending to drown at the bottom of the pool to summon the rescuebot.

[frobozz]

Yes, and that's the other thing.

To reliably distinguish between drowning and breath-holding, you need to wait (or have more context)

This device (which has to come out through a secret underwater hatch and propel itself amongst people) cannot be faster than a lifeguard who has spotted the questionable situation and put themselves close to it before making a decision.

[Humdeee]

I can see the famous kissing scene from Squints in "The Sandlot" movie remade once again.