Hacker News Reader

Comment by joshuahedlund | Hacker News Reader

> The latest Cruise incident occurred Thursday night when a Cruise robotaxi and an emergency vehicle crashed and left a passenger injured. Cruise said in a social media post that one of its self-driving Chevy Bolt EVs entered an intersection on a green traffic light at Polk and Turk streets when it was struck by an emergency vehicle that appeared to be en route to an emergency scene.

Noticing and responding to emergency sounds/signals in the road, distant but approaching, is yet another subtly complex aspect of human driving I hadn’t thought of but which seems obvious in retrospect. Seems solvable in theory but also involves a lot of uncertainty and contingencies. Then I suppose the cone-blocking hackers will start playing siren recordings to confuse the robotaxis…

View on HN

Cruise released a statement which said the car did hear the siren when it was louder than background noise. No indication how close the siren was for that to happen, but it sounds pretty definitive that audio sensors are already part of the kit.

How good are those sensors relative to human ears? Human ears are still freakishly effective at filtering out background noise.

A small nit (sorry): it’s probably not a matter of sensor quality (decent quality microphones are not very expensive), but rather the neural networks that process the audio signal.

It’s a blend of both. The human ear has a vastly more complex structure than a membrane and a sensor.

We have complex flexible shapes to channel audio. Alongside an extremely complex set of hairs for hearing.

That should be part of an assessment that government does before allowing those vehicles onto our streets.

I don't think you need freakishly good filtering to find something as pure as a siren.

It’s almost a harder problem - I can regularly hear sirens in my town the moment they leave the fire station until they arrive at their destination - more than a mile away in each case.

A siren does not have to be lauder than background noise to be hearable.

> Then I suppose the cone-blocking hackers will start playing siren recordings to confuse the robotaxis…

Ummm, that would even confuse normal drivers, and I hope that it would wind you up in jail.

Those siren sounds get played on the radio sometimes, and I really wish that it was illegal to do so.

In BobbyBroccoli's "The man who tried to fake an element", the title sequence music has a police siren play at one point. Every time I watch the video, I think there's an real-life siren outside.

Cone blocking isn't hacking. It's simple criminal behavior.

Understanding a system well enough to manipulate it in ways that the owner doesn't want seems at least in the same category of phone phreaker behavior.

Which was also criminal behavior, but the point is that I don't think it automatically disqualifies it from the category.

Exactly. Some criminal acts are so simple even an animal could do them - destroy something.

Others are so complex or use the software against itself that they clearly are hacking; even if they remain criminal.

Is putting a cone on a vehicle to cause traffic jams really so complex? Perhaps the first person to do it was a bit clever, but anyone imitating it really isn't.

I think it depends. To compare it back to the phone phreaker example, putting up a single cone that blocks something is roughly the equivalent of cutting a phone line. That is to say, probably not hacking.

But what about when you add a level of complexity and creativity? Say, putting up a series of cones that trap the car in an endless circle?

I don't really think the semantics matter too much. Sure, call the circle of cones a hack. The point is, it's criminal & likely causes a degree of chaos well beyond a simple prank.

That did not seem to be the point of the original post that I responded to.

Specifically, it seemed to imply that anything that was criminal could not be hacking.

The legal status of something, seems completely orthogonal to hacking to me.

That wasn't my point in the original post. Obviously overlap between those two things is possible.

Then I completely misread the post. In my defense I went back and re-read it, and it still looks that way. But, y'know, tone and text and all that. I could always be wrong.

What was the point then?

The point was that it's a stupid and harmful crime. Calling it a hack makes it sound like a clever harmless prank. The point is about the characterization of it, not about what hacks can and cannot be.

which law is being broken?

Good question & to be honest I'm not 100% sure, though I bet there are many laws vague enough to capture it.

If it causes extreme traffic jams, perhaps there's some kind of public nuisance law that will cover it? Is interference with public infrastructure a crime?

Do you actually think this type of behavior will simply be tolerated?

as a vehicle operator: if someone put a cone on my bonnet, then I just sat there blocking the the highway I'd expect to be the one fined

quite why this is different because an unfit for purpose driverless vehicle is involved, I don't know

the company operating the vehicle should be the one fined for blocking the highway

They don't put the cone on the bonnet for no reason. They do it because they know what the likely outcome will be.

In a similar way, if you booby trap a vehicle such that it might surprise a human driver in some way whilst they are driving, causing them to potentially have an accident, the crime you are committing is not simply that you're surprising someone by tampering with their property, but you might be doing it with a specific intent to cause an accident.

> the crime you are committing is not simply that you're surprising someone by tampering with their property, but you might be doing it with a specific intent to cause an accident.

this is nice rhetoric and all but doesn't apply at all in this situation

nothing has been tampered with: the cone can be removed and having done so the vehicle will be indistinguishable from before it was placed (no property damage)

the "attack" keeps the vehicle in its safest possible state (that in which it can't move) so there is no intent to cause an accident

and you still have yet to point out which law is being broken (I suspect because there isn't one), so it is a crime?

however the company may be able to go after you in a civil court for lost revenue/reputation

You'd be doing it so the vehicle won't move, and if the vehicle doesn't move it will cause traffic. If your argument is even slightly correct (that there would be any problem prosecuting such behavior) then it at least won't be in the future. Obviously trying to trick a car into causing traffic chaos is not going to be an accepted behavior.

Maybe we should try other signaling methods (radio?) for emergency vehicles that’s easier to process for computers. That may be end up being useful for human drivers too ultimately.

Why would we do that?

Because nowadays cars are computers on wheels with lots of sensors, so we don’t need to only rely on human ears and brains like we had to in 1950s.

I think modernizing traffic signals will end up making transportation more efficient and safer. (Just imagine knowing the timing of the next 5 traffic lights on your path.)

This would make roads undrivable by human drivers (unless augmented by modern tech). Probably exactly what you had in mind, but I don't want this future (at least not now)

> Noticing and responding to emergency sounds/signals in the road…

How do deaf drivers respond to sirens from emergency vehicles?

Speaking as someone who drives emergency vehicles with flashing lights and sirens - it's amazing how many people presumably with average hearing and average vision cannot see a fire truck until it's been tailing them for a while.

As a frequent lights and sirens passenger I'm convinced that they cause such a huge adrenaline spike that surrounding drivers act completely irrationally. Drivers will see us and then slam on the gas to dart in front or sometimes just stop in front of us. So many close calls where our engine almost gets hit by a human driver who seems completely committed to hitting us. Or we'll be turning onto a street (sometimes we have to swing wide to take a tight corner) and there will be a driver trying to squeeze by and making the same turn as us.

If you're in the US, please pass feedback to whomever it makes sense that the signals are just too loud to the point of being dangerous. Thank you :-)

(I get the impression these are just a tad quieter in Europe)

You’ve not been a city long if you haven’t seen an emergency vehicle have to get on the PA to start yelling at a vehicle in person.

Rumbler sirens can be felt, in addition to the tonal difference: https://en.m.wikipedia.org/wiki/Rumbler_siren

Flashing lights

There are many instances in dense cities where the streets are too tight between buildings or busy with buses and cars to see down the other road at an intersection.

Japan gets around this by having ambulances at least slow down and look both ways for each red light they are driving through. Doesn’t seem like a huge ask.

Emergency vehicles here do the same thing, but there’s a human element that may be hard to implement- the drivers assume that a vehicle has “seen” them if it stops. But an automatic car (and some drivered cars, too) will then start up again and drive into the side of the truck.

But the biggest reason they slow down is that if two emergency vehicles are responding from different roads - they can hit each other - so much so there are multiple examples of it on YouTube.

That’s why they switch sound approaching an intersection and slow down. It actually helps them if there are cars stopped at the cross streets, running interference.

The big one too is noticing everyone else slow down so you start trying to figure out why and look around more.

Quite a while ago, I learned the best way to follow the ball or puck at a sporting match. You don't try to look for the little object itself, you just observe what all the relevant players are doing, where they're looking, where they're running. You figure it out really quick with them helping!

Likewise, if I am waiting for a bus or something, I am really alert and on guard if I'm alone at the stop, because that bus will blow right by. Once there are a few other passengers around, I kick back and just watch them, because unless they're all impatient and pacing, their attitude and behavior will indicate when the bus is visibly arriving.

SDCs that coldly and analytically examine the traffic around them will probably never get the human nuances that we can pick up from second-and-third-order clues. For example, if I'm paranoid about whether a car's going to turn into me in an intersection, I don't watch their turn signal or lateral motion, I watch their front tire, because that's the first indicator of a turn. Is that sort of thing baked into AI? If several people are stopped at an intersection, eye contact and gestures are often essential in negotiating a solution. AI? Same with pedestrians. Make eye contact, signal, work something out. I'll often stop on a corner and wave cars through rather than insist on my right-of-way. Of course the motorist also has to decide if that's safe to do, and sometimes you can fool a motorist into making an unsafe move just by signaling your permission...

Most drivers aren’t deaf, and many people who are hard of hearing can at least hear a siren approaching.

If all drivers were completely deaf it would indeed be a problem.

Honestly, looking at recent news reports in the Chronicle, I don't think cone-blocking hackers need to do anything more right now. I mean, why cause trouble for someone if they're already slipping up in this way?

In fact, if they kept pushing it gives an easy excuse - it’s the coners! Even if not completely true.

You have to know when to stop and let the cards fall where they may.

Emergency vehicles in many areas have a strobe light that sends out a particular pattern of flashes, allowing them to turn traffic lights red as they approach. Seems like a fairly obvious step would be to mandate that self-driving cars recognize those strobes.

If the light is already turning red that might be the more obvious indicator? Also those infrared LEDs are kinda angled upwards towards the traffic lights and might not be as visible from the road

And the training data for that situation is hundreds/thousands of times less plentiful than training data for other scenarios.

Reply on news.ycombinator.com