Hacker News Reader

I am actually slightly surprised that road signs haven't been updated to include something like a "barcode" or similar that would make it very easy for the machine to identify what sign it is. You could "paint" over the entire sign with a variation in the infra-red reflectivity with a QR code, for example.

Then the computer would be better at recognizing a sign than a human.

Back in the late 2000’s I wanted to start a company that broadcasted the current speed limit near signs. Like a BT beacon, sorta, but obvs not BT. It turns out this requires too much coordination. Specifying a protocol, getting manufacturers to build it in, lawmakers to pay for it, sign makers to embed it, etc.

The tech is the easy part.

Isn't this solved by GPS and maps with speed limit data? Google maps gives me nonsense speed limits in a couple of places but mostly correct.

No, because the lower speed limits posted on flashing signs for construction are an important local case.

In my experience, Google just thinks those are traffic jams. I've seen Maps 'know' about road construction speed limits, but it isn't usual.

Good point. Also, M25 (London orbital motorway) has large sections with variable speed limit. Google maps navigation just doesn't show a limit for those areas, since it does not have timely data.

Also the 20mph boroughs really confuse the systems (cars and satnav), since they very often say 30 in 20 and also 20 in a 30. The various limits are, admittedly, not well signed, and I suspect that's deliberate to keep drivers cautious.

One problem is "map matching". That is, given your lat/lng (or, more realistically, a time-series of lat/lngs) and a map, figure out which road you're on.

The thing is that most consumer devices with GPS (like phones and even standalone GPS navigators) have a lot of internal "magic" to hide the raw noise in location fixes from the GPS signal.

For example, certain Garmin devices (intended to be used off road) will snap your location to an Interstate if you are going more than 55 mph and heading roughly in the direction of the road (which is evident when you run into traffic and your location keeps switching between the frontage road and expressway--the error distribution of the noise doesn't have mean=0!).

Out of curiosity, how did the tech deal with things like frontage roads that have much lower speed limits than the highway right next to them.

Basically, they haven't, or at least they hadn't 4 years ago when the car I have with it was made.

Diving along a free flowing 50 and having it decide it's now a 30 because it saw a sideroad and drop a gear is semi-regular. The second most dangerous thing it does after having a touchscreen.

It also manages to read 45 sometimes but I have literally never seen a 45 sign.

Radio waves can be directional. So they were pointed at the road they influenced, not the highway.

If embedded in the sign, what if the sign itself is rotated or damaged in some way?

Our prototype just screwed onto the sign post and could be aimed that way. We never added it to the prototype, but having an accelerometer added to detect movement and shutting down/calling home would help with damage or vandalism.

This would probably be easy to exploit. Someone could drop a bunch of random invisible stop signs.

to be fair someone could also go toss a steel bar on some railroad tracks or drop rocks from the top of a highway overpass.

Right, but that's rather obvious.

It's true that people aren't often looking for fatal pranks, but you could do more benign like try to get people pulled over by causing them to speed unknowingly.

Someone could really just weld a steel bar onto a railroad track out in the wilderness and derail a train??

Yes? No welding required, just a few thumbscrews, if you use one of these:

https://westernsafety.com/nolan-portable-derail-pd-2/

On sale now! Or pick this one for bigger and faster trains:

https://www.aldonco.com/store/c/127-SaberTooth-Portable-Dera...

There are lots of tools, weapons, chemicals, and other things that a person could do that would be bad. It doesn't take much creativity, just a callous disregard for other people. But in contrast to the fearmongering and crime you'll see in the media, consider it a bit of good news that in spite of how incredibly easy it is to intentionally cause harm to someone else, billions and billions of people choose not to do that every single hour of every single day.

I mean, you now have told a lot of people that you can just get a tool to derail a train easily, that I'd guess most (I didn't) don't even know it exists.

I think that's GP's point: derailing a train is just a search engine away, and that's only scary when we think about it, which we almost never do.

Which in turn is because our emotions around people setting out to do that, are calibrated around the fact that it basically doesn't happen outside of war conditions.

I googled how to do it just for lulz and got youtube video with someone showing one... and some russian/ukrainian looking name in comments asking about it and asking where to get one, from like a month ago

https://www.kyivpost.com/russias-war/anti-putin-russian-rebe...

I wonder whether it's related. Probably not but train derailments are weirdly relevant to current world state...

But back to the topic... I don't think anything should be censored and am of opinion freedom of speech just need to be taken wholecloth, good with the bad, but you gotta wonder how much easier it is now.

AFAIK, at least in Russia there are signaling systems that will interpret a metal object connecting the rails as "another train is on the track" and inform the train driver.

Not sure whether those systems are also installed in the wilderness though. But it seems kinda likely that anybody thinking "I should derail a train for fun" might not bother driving into the wilderness when there's an alluring, convenient, seemingly unobserved stretch of the railway available nearby.

Edit: another thing that comes to mind is that you'd also have to find a low-visibility part of the railway, even in the wilderness — because otherwise the driver will just notice the obstacle and stop the train.

You don’t even need to weld it, you just need to leave it somewhere in the way and striking it at sufficient speed will result it in bludgeoning through a car or worse.

An improperly stored rail from a maintenance project derailed a NYC subway train in 2017: https://apnews.com/article/a6a01e5928f9478bab0d33f03260870e

Yes. Railroad tracks in the middle of nowhere are usually freely accessible. Why bother putting a fence if no one's around? Even at railroad crossings in cities, one could cause a some collateral damage by welding it right next to the crossing.

Its hard to derail a train.

https://youtube.com/watch?v=agznZBiK_Bs

Are you genuinely asking about the feasibility? If so, then: yes.

Tag the coded signs on a GPS map. If the detected "sign" is not on the GPS map the system can label it as "probably malicious".

Then why use the signs at all rather than just use this gps based road rule database

Because not all cars are self-driving or have the ability to access this hypothetical database. That will remain the case for likely the next 30 or more years.

Right, but the cars that are scanning signs and will have the ability to read said code are the ones being suggested that they should consult this database to avoid errors. If they're consulting it to avoid errors, they could just use it directly

Now you have two sources of truth for autonomous and non-autonomous cars. You end up with two groups playing the same game by different rules on the same board simultaneously.

Whether or not that's a problem in practice is not clear to me. Among others, taxis, construction vehicles, emergency vehicles, garbage and delivery trucks, municipal busses, school buses, bicycles, motorcycles, and vehicles in funeral processions all play by different rules than the average motorist.

This mostly doesn't cause disasters. Of course they're visibly different for the most part and make up a comparatively small percentage of traffic.

Ride hailing vehicles might be an interesting point of comparison: they're visibly similar to normal vehicles, comparatively common, and play by something resembling taxi rules.

honestly if we're going this far we may as well embed a speed system in the road itself and have the signs for redundancy.

I can see potential problems with temporary signs, such as those put up for road works, around accidents or for police enforcement.

Plenty of places have no reliable GPS reception.

That's why we need to sign the signs.

And use a blockchain to track every interaction between passing cars and the sign. Which makes speeding tickets all that easier.

https://xkcd.com/1958/

Signs are already reflective and pretty standardized. I don't think machine-readable codes are really necessary for this.

One advantage of something machine-readable, like a QR code, would be to incorporate additional information (more than could be easily consumed by a human).

For example, a speed limit sign could include the name of the road it controls, which would help reduce instances of cars suddenly slowing to 35 mph on the freeway when passing a sign on the frontage road.

It might also be useful for cases of (this is US-centric) yellow "advisory" speeds posted for, e.g., sharp curves. The machine-readable aspect might indicate how long the advisory speed applies (next 1000 ft) and why (sharp curve, hidden intersection, etc.)

If you just want to feed computers just put a database with every sign location and data.

That way even faking or destroying sign would be ineffective.

Would have to be updated during temporary situations (roadwork etc) tho

> That way even faking or destroying sign would be ineffective

In theory you could add an HMAC (or similar) signature to the QR code. It would be bigger, but if cars maintained a DB of state DOT public keys, you could mitigate this somewhat.

Then you’d also have to worry about duplicate codes, but at least they wouldn’t be “fake”. Maybe even add lat/long coordinates to the code… again, more data, more pixels required, but it would make sure the signs are roughly in the right place. Which doesn’t seem all that feasible to require construction crews to put the right sign in the right place.

Logistically this doesn’t seem all that promising.

Just give the road crew a phone app that can read the sign.

Given the phone's camera, compass and GPS, validating that the sign has been installed as specified (location, facing direction) is almost trivial.

We can barely just keep a road painted. There’s no budget for wholesale replacement of signs, which often have replacement spans in terms of decades.

Or they could have something like an IR beacon, that way cars can recognize them without relying on computer vision and nothing that looks like a road sign can confuse the system.

They have specific shapes and colors, which probably work the same way

How about blinding the (meat) driver with a low-powered laser? If you want to murder someone, there are many options.

I wouldn't bet on "no evidence" for the autonomous vehicle. It may record telemetry and video.

> It may record telemetry and video.

For the novel, probably neither of the above is recorded in high enough resolution.

Thinking of subliminal messages inserted into old movies here, just flashing something up for a hundredth of a second, doing it enough that a system detects something... but not enough that video filmed at 30fps really shows it, and that telemetry doesn't really capture it.

All the telemetry would show is "car thought X", and the video would appear to be clean. What isn't known from is "why" car thought something.

We have systems log stuff when there's less at stake. You'd have to really go out of your way to make this an unsolvable mystery murder.

That's a shame, I was thinking a modern Crow Road (Iain Banks) with it being an accident that only later you pieced together that a person was in fact a suspect because they'd had these skills and could've done this far out thing.

Old movies are projected at 24fps so subliminal messages like you mention would be noticeable from a 30fps camera.

Most interesting to think about are the stickers[1] that look random but actually trigger some kind of adversarial behaviour, a bit like how tiny human-imperceptible changes to an image can completely flip an image categorisation network's guess.

Perhaps most fun to think about in a Dan Brown-style NSA capabilities combined with Peter Watts' Blindsight's near-total-imformation perceptual-cognitive shenanigans kind of way.

[1]: https://medium.com/self-driving-cars/adversarial-traffic-sig...

Add the "autonomous car trapped by a circle of salt" to your repertoire if you haven't seen it already: https://www.youtube.com/watch?v=thuN2HD6m2s

Except in the self driving computer's recordings of its inputs...

First, display an uproad pattern crafted to cause a buffer overflow. 0wn the autpilot, trash recording. Bonus points for taking over the entertainment system, cueing Death Grips "Guillotine" at max volume as the car leaves the road.

Humans also rely on visual information for driving and can easily be tricked. For instance, I don't think there is any security mechanism to prevent the placement of a battery powered green light on top of a red traffic light. In that case, an autonomous car's logs could actually help in discovering the tampering.

No, but the reason there are three (at least) lights is that the designers were well aware that colorblindness exists and the position of the light is also significant.

It's still kinda shit design to be honest.

Could have both color and shape. Make red hexagon, yellow square for example. or really any recognizable shape

I'm sure that manufacturability is a concern.

I'm sure it's no harder than making one in shape of arrow lmao

Interfering with a traffic signal is a crime in many jurisdictions.

> Plot for a future novel: Project light onto surfaces from a moving vehicle to make another car behave in a way that kills the occupants. No evidence can be found at the scene as to why the car killed the occupants.

Aside, well, the video the car recorded...

Autonomous Trap 001 is another: https://jamesbridle.com/works/autonomous-trap-001

I was actually on a 40mph road earlier today when I passed a side street with a 20mph sign that had been rotated to face the main road. My car was then insisting it was 20 for the rest of the road.

>Plot for a future novel: Project light onto surfaces from a moving vehicle to make another car behave in a way that kills the occupants. No evidence can be found at the scene as to why the car killed the occupants.

Problem with this is that self-driving cars like teslas will probably have cameras all over them

We are lucky that most people aren’t murderers: https://xkcd.com/1958/