Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

How would a malfunctioning AI decide it's safe to stop?


It wouldn't decide whether it's safe to stop. A separate E-stop processor would have a preprogrammed stop routine that stops the vehicle as fast as it can stop safely, which is to say, without skidding and without making rear-end collisions inevitable if you're on the freeway.


> and without making rear-end collisions inevitable if you're on the freeway.

That means it has to decide whether it's safe to stop.

Stopping safely is no less complicated than the self-driving software in general. Sure you can have backup driver software, but it's just going to be the same software.


> but it's just going to be the same software

It does not have to be the same software. Its task is much simpler, and many situations where the normal software has to look for an optimal resolution can be handled by activating the hazard lights, stopping the car, and telling the passengers that the emergency stop is complete.


People talk a bunch of stuff about malfunctioning, bugged software and so on. We can assume, the car was just trapped in some infinite loop so it'll never reach it's goal for some reason but also wasn't able to realize that and call the routine for that. Fine.

It's not a dangerous situation for the passenger but it definitely something that makes you feel very uncomfortable which could lead to panic because you're trapped. The doors are locked and even if they weren't you cannot simply exit the car. That's okay. Trains don't allow this either to open the doors on a track section. You need a supervisor to do that. In case of an emergency you can still unlock the doors mechanically. So after pulling the emergency brake you can still do that. It would be interesting how these cars behave in case of a fire inside or something which leads me to my final thought:

In case of issues on or with a ride in a theme park, there are usually operators. So in case of a technical malfunction there's an emergency stop which should make disarm systems in a safe manner and bring the ride into a safe position or at least make sure it stops safely. Of course it's not possible to trigger an e-stop as passenger but an operator can do that in case you can somehow make clear that it's needed right now.

The situation in the video is a very bad experience because a phone call from the passengers device is needed and he was asked to do something in the app. Probably to identify the car and enable the remote control. Elevators handle this much better. There's an emergency system in the cabin which transmits all important information to the service center, so it's immediately clear where the emergency call originated from. An app on a smartphone should never replace this installation!


It happened not to be a dangerous situation for this passenger, but that is not always the case. See https://news.ycombinator.com/item?id=42646148 for a list of examples.

When I pulled the emergency stop cord on the train, I also pulled the emergency door opening lever. The doors opened, on a track section. I've used an emergency door opening lever more recently when the train had already stopped and we were being instructed to exit because of a mechanical problem (possibly a fire), but that was at a station.

Elevators, in my experience, also universally have E-stop buttons or switches.


Add a couple of pedestrians to that parking lot, as in a mall, and that situation turns dangerous quickly.


No, it doesn't. It just needs to brake slightly less hard than the antilock braking system is capable of. There is no complexity there. The only complexity is in ensuring that the E-stop processor can reliably override the rest of the car's control systems.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: