Hacker News Reader

Comment by rjmunro | Hacker News Reader

"Truck reverses into stationary object" is not really news.

I think there are too many possible risks of programming the vehicle to do anything except stop dead in an emergency situation - you rapidly get into very complex programming with all kinds of failure modes - e.g. what if the problem was a faulty sensor? The vehicle might try to avoid a non-existent threat by crashing into something else. Stopping is the safe thing to do. If you want to react, you rapidly have to make lots of moral decisions like (The Trolley Problem).

Briefly sounding a horn in an unexpected emergency stop situation is probably a good idea, particularly if it's a white noise type thing, rather than a siren, so that other road users can localise it's source quickly.

View on HN

A horn needs to be a horn, add a white noise generator too, but the recognised noise and the one people are "programmed" to react to is a car horn.

Aside: emergency vehicles in the UK have stopped using the broad-band noise sirens AFAICT, I don't know why but the reason may be pertinent to any attempts to use it here? Actually, they often ride without the siren at all, which is completely ridiculous to me; they put it on at the junction meaning there's no time to get out of the way, if they used it continually you can hear them coming.

They're trained to look ahead, anticipate and adapt to the road conditions, judging when they need the siren and when they don't. This is important in built-up areas, because there's no gain in waking up people at 2am when you didn't actually need a siren to proceed safely.

When they're putting it on at a junction and you're already in it, they're not expecting you to magically disappear out of their way. They're expecting you to look at where they're going and stay out of their path. This might involve proceeding as you were going to in order to clear it for them. What they really want is to stop anybody else from entering it until they've got through.

Here, UK, they don't put it on at all during the day, travelling through built up areas, particularly Ambulances it seems.

When the sirens are used it's great because you can tell what road they're coming along (after a while, it was much better with the "squelch" sirens) and anticipate in good time.

Agreed, waking everyone at 2am is not needed generally, but howabout alerting people at 3pm.

You appear to have inside knowledge here, presumably there's been some change in policy on this in the last - I don't know - 5 years, where it's been chosen during normal daytime callouts not to use sirens if at all possible?

The argument for self-driving cars is that they will be better than a human driver. A human driver would backup, sound the horn, or both. If a self-driving car doesn't attempt these things, something is deficient.

A human driver can shout and gesticulate which will enjoin passers-by to engage with the situation - waving and shouting too. I've seen several slow speed (eg parking) accidents avoided this way. Perhaps these vehicles need a "I'm in danger" external indicator of some sort too?

You mean like the hazard (sic!) lights and the horn? Out where I live, having those functional are amongst the hard requirements for street-legal vehicles.

No it's a separate, more primal, layer of hazard indication that seems to shortcut the Bystander Effect and enlist aid. It can be initiated by non-drivers, pedestrians, and cyclists too.

Alternatively - these are just growing pains until the human driving the truck is replaced by an algorithm.

So basically, self driving cars will be worse than human-driven cars until ALL cars are self driving cars? This is the first fully autonomous version of "No True Scotsman" that I've seen!

I don't think it is an instance of "No True Scotsman" to say that there are situations that self-driving cars may never be as good as humans at, but that those situations only occur because of the unpredictability of human drivers.

Self-driving cars don't have to be better than humans at everything in order to be objectively better drivers than humans. In 2010, 32,999 people died in car crashes in the US [0]. If self-driving cars are on-balance 1% better than human beings at driving, then 329 lives are saved.

[0]: https://en.wikipedia.org/wiki/List_of_motor_vehicle_deaths_i...

This is the reason I believe that self driving thing is doomed. Significant amount of humans will refuse to stop driving. Fair amount of them will make life of self driving things unbearable. Sizable part - on porpoise. Just to check how good a computer is.

Seriously speaking from perspective of a truck driver.

The reasons truck driver backed into thing are: He couldn’t see it. It wasn’t there when he started.

Horn and flash of lights would stop a human driver. When I back up a truck I use all my senses, since I can’t see much. Easiest fix for the robot thingy - add horn and lights. (And wait until all residents start complaining)

People do die of old age, y'know. And long before that, they stop driving because their eyes become too bad and their reactions become too slow.

All you have to do to cement a driverless-car hegemony is to stop issuing new driver's licenses. 50 years later: no more drivers.

I read a sentiment that I found myself nodding to: people will stop driving when it’s convenient for them.

Some people love driving, and some people love riding horses, but you don’t see many horses on city streets today.

The argument for self-driving cars is that they will be no worse than a human driver.

I'd love to show you the video, but it's been removed. It shows a Starship technologies sidewalk delivery robot attempting to cross the road, when a car making a right turn through the crosswalk hits it, in a low speed collision. The interesting thing is that the little robot attempted to make an evasive maneuver by stopping and quickly backing up to avoid the car. It ultimately failed, but at least it tried.

So Navya, the French company who has deployed several of these slow moving autonomous shuttle busses does not appear to have such sophisticated behavioral algorithms, in spite of entertaining ambitions of building proper robotaxis.

A small robot is a lot less dangerous when backing in response to an emergency than a car. The small robot can't accidentally drive over and crush something.

> I think there are too many possible risks of programming the vehicle to do anything except stop dead in an emergency situation - you rapidly get into very complex programming with all kinds of failure modes

Perhaps that limitation should disqualify AI from driving cars.

AI only needs to be better than people. I've seen a person drive the wrong way round a roundabout here. The bar is very low.

AI will need to be better than a good human driver, so these kinds of comparisons are not helpful. A good human driver would back up out of the way of the truck and/or sound the horn. If this is too hard for AI, then it is not ready.

AI needs to be better than the average human driver to drastically reduce accidents and deaths. We agree on the horn though, it should know when to honk at things.

A SDC needs to be better than humans on average to be a safer choice. It doesn't need to be perfect. Now, getting humans to acknowledge that and embrace that is the real problem.

Particularly since most drivers think they're above average. A car which is just better than the average driver might be a tough sell for them.

Practically it needs to be perfect, because whoever makes the self-driving car will be liable every time it kills someone. And there's going to be millions of these things.

Right. It's easy to Monday Morning Quarterback these things when you don't have all the details.

To name just two potential issues that the "bus should have moved" people are ignoring:

1) What if there were pedestrians around?

2) What if moving the bus created an incursion into a stream of fast-moving traffic without warning?

Both of those could result in consequences far worse than the (apparently minor) fender-bender that actually occurred.

I agree that sounding the horn would be a good idea.

Reply on news.ycombinator.com