Hacker News Reader

Comment by ckmiller | Hacker News Reader

I competed in the DARPA Urban Challenge, back in 2007. While we were watching the finalists trying to sort out the world's first fully-automated traffic jam, the conversation turned to how easy it is to manipulate these vehicles.

Imagine a remote stretch of highway frequented by automated 18-wheelers. All that's required to bring the truck to a screeching halt is a bedsheet and some decent timing, at which point the vehicle has no way to prevent a robbery. The truck could put in a remote distress call, but it will still be some time before a human can get there. It's a new era for railroad heists!

We're no closer to solving this problem than in 2007; everybody is still trying to manage the long tail of merely safe driving. Handling humans in adversarial situations like the above is still completely off the map.

View on HN

Why does this sort of critique keep coming up in HN threads on autonomous vehicles? have autonomous cars really gotten so good that "what if violent criminals try to attack it" is the only criticism left? I'm pretty sure not hitting cyclists is still a much larger problem than accidentally wandering into the plot of an action movie. less interesting, i guess.

theft is an insurance problem. If there's no human driver to get killed, it doesn't really matter if the truck gets robbed.

have autonomous cars really gotten so good that "what if violent criminals try to attack it" is the only criticism left?

No, they haven't. This is (by far) not the only criticism, but it's a valid concern which has literally zero attention to addressing it.

I'm pretty sure not hitting cyclists is still a much larger problem...

Probably, and that one still isn't fixed either.

What makes you think no one is considering how to address spoofing? It seems likely to me that many of these companies have considered these cases in far more depth than you or I have.

You're correct, autonomous vehicles are absolutely not to the point where highway robbery is a prime concern. But even if they were, the spoofing issue isn't just another tough problem to solve. It's a potential arms race.

Faking out machine learning systems is rapidly progressing from a few "fun proof-of-concept" examples to a serious area of study, and we've already seen it (gently) applied to autonomous vehicles [1] (ignore the overblown headline, it's just a piece of tape on a sign).

[1] https://www.technologyreview.com/s/615244/hackers-can-trick-...

> It's a potential arms race

but it's currently solved by one guy sitting in a truck, so it's not really a potential arms race that can escalate too far before you fall back to the one guy sitting in a truck.

worst case scenario, you replace a driver with an actual security guard who is trained as a security guard and can focus on security instead of driving.

A human solution does seem like the real solution to the current shortcomings of automated driving but security guard seems like too narrow of a role. If you could have a shepard whose job was to sit in a lead vehicle that identifies, tags, and issues orders for avoiding obstacles in real-time for the automated herd trucks then the problem is 99% finished as far as creating a new business model goes. City streets are terrible so leave that problem to the futurists, one guy herding flocks of five or ten trucks to depots outside the city would already represent massive savings.

This is how automation actually progresses. It's not bam, one day, truckers are replaced with AI that's equally competant at driving. Rather, day by day, the amount of freight moved, per unit cost/human effort, slowly decreases.

Much of this efficiency has already occured in terms of routing optimization.

worst case scenario, you replace a driver with an actual security guard

Or even just do that for a certain percentage of trips. (See "Margin of Profit" by Poul Anderson).

In a self driving world of well behaved traffic, police and highway patrol are going to need some way to maintain their jobs.

One large difference is that a lot of people hold back from theft purely because there's a human involved. Doing illegal things to human beings is a whole higher tier of scary, compared to doing those same illegal things to robots.

Consider the difference between robbery and shoplifting. One is the domain of rare sociopaths; the other is common-enough to be almost a subculture among teenagers in certain areas.

Robbing a robot is, in some sense, just shoplifting (or, more accurately, warehouse theft) performed against a moving target. It's not "the plot of an action movie"; it's some unethical people's lazy Sunday afternoon. I'd expect it to be far more commonplace, once it is possible at all.

And, as well, keep in mind that, the more predictable something is, the more you can automate attacks on it. It's a lot easier to have a worm going around stealing Bitcoin out of people's digital wallets on their PCs, than it is to steal real wallets. Likewise, it's a lot easier to create some kind of drone that steals from other robots, than it is to create a drone that steals from human-driven vehicles. In combination with the property of people feeling much less compunction against attacking robots in the first place, I'd bet that you'll see e.g. delivery drones having their cargos hijacked by pirate drones, with increasing frequency. Autonomous trucks are just a larger-scale variant of the same.

Pirate drones? Really? In the real world autonomous systems get hacked and bevome the "pirate" drones themselves. It's quite boring, I know.

Is a warehouse on wheels absolutely festooned with cameras less secure than a warehouse in the middle of nowhere?

Yes, people don’t leave warehouses without guards if there is anything of great value in them that’s easily to walk away with.

If you're not trusting stuff to sit static without a human near by you probably wouldn't let it hurtle down the road with the same.

On top of that, I doubt any human trucker is defending their load against two humans with shotguns, and that doesn't happen today enough to make the news. If anything else, you can make it so the vehicle isn't drivable by a human (no interface) and make the cargo door super secure.

> On top of that, I doubt any human trucker is defending their load against two humans with shotguns, and that doesn't happen today enough to make the news.

True, but if you aren't assaulting / endangering a driver, then the crime is lower-risk.

But if you are robbing an autonomous vehicle, you're probably going to be recorded doing it, and fleeing, which raises your risk because it leaves a pretty strong evidence trail. Yes, ski masks can be used to obscure your face, but car makes can be tracked if the getaway car has no plates.

If this becomes a frequent enough issue to matter economically, then perhaps we should start looking into what incentivizes stuff like that in the first place, and deal with it.

A getaway car isn’t needed, you just amble away and walk two miles the other direction.

If you just robbed a truck, are you going to carry all the contents on your back? For two miles?

Up the hill! Both ways!

A $100 foldable shopping wagon can carry $100k of iPhones trivially.

The thiefs use a laser to blind the cameras. Or a mirror to reflect the sun. Or spray paint. Or a smokescreen. Or other solutions. Easy.

> The thiefs use a laser to blind the cameras. Or a mirror to reflect the sun. Or spray paint. Or a smokescreen. Or other solutions. Easy.

None of these would work, except maybe "other solutions".

A self-driving vehicle will have far, far more sensing than any of today's trucks. Someone will know you're robbing them the moment you start, know where you are, be able to watch the entire event as it happens, and every ounce of it will be recorded.

And none of that will matter because they will just have a video of someone in a ski mask robbing a truck and walking away.

Robbing a truck of what? Trucks are full of heavy stuff that's not particularly valuable. Were it otherwise it wouldn't get shipped in an unguarded truck.

iPhones are not shipped in a guarded truck.

Walk away with 20 tons of stolen goods? How are you going to accomplish that? Why not do something interesting like reprogram a whole fleet of cargo trucks to deliver the cargo to the wrong destination?

People would steal from trucks significantly more frequently if there was no risk of harm like there is with the shotgun scenario.

I had two bikes stolen that had locks at them (at different times) from my apartment in Mountain View. Theft is rampant and removing the threat of a real driver pushing back completely changes the calculus for trucks.

It, to me, fits in with a wider implicit value on HN that property is worth more than people.

If I'm a truck driver who someone tries to rob ..sure just take it. Bit my stuff, that's why I have insurance. Who cares if a truck gets robbed if it protects a human life?

If we’re assuming these vehicles are FULLY autonomous, a simple fix for this is to add a manual override that allows the driver to drive for themselves. Then the heist is basically as effective as it’d be today. (To be honest, I’d be surprised if such a feature isn’t required by law).

Though off-topic, the conversation of vehicle-feature-defect-related heists reminds me of Jaime Zapata, a US agent killed in Mexico after his SUV doors automatically unlocked when parked. (Though it’s unclear if locks would have made any difference in the outcome).

https://en.m.wikipedia.org/wiki/Jaime_Zapata

You can already do this to humans pretty much just as easily, I just need some orange cones

Then what? You have 20 palettes of dog chew toys or asian pears....? Some of the goods have serial/tracking numbers.

The reason people don't rob 18 wheelers or trains is because it's not a good criminal enterprise

I have not thought much about robbing trucks before, but I disagree this is the same as a manned truck. I think it's much more like the difference between robbing a house that is empty and a house while the owners are home. Most burglaries happen when nobody is home. Why would trucks be any different?

Besides that, it is (likely) easier to trick an unmanned truck into stopping as OP mentioned. A person would easily recognize what's happening and it would be easier for him to just drive around the obstacle. It's not as easy to create an AI that can "think" around things like that. Besides that, I hear truck drivers are often armed. I would guess that punishment for property crimes is less severe than armed robbery against a person, but I don't actually know.

In the grand scheme of things though, I think this is a negligible concern compared to all of the other problems self-driving trucks have to solve first.

The unmanned truck would likely be _crammed_ with cameras and other surveillance tools like hidden gps senders in the containers.

It might initially be easier but I'm not sure if you would get very far with the stolen goods.

I used to work for a company that made long-life self-contained GPS locators, for commercial asset tracking. I remember how much we chortled ironically when a shipment of our devices was literally stolen by a hapless meth addict who noticed keys left in the ignition.

Everyone was laughing because "ha, that idiot stole a thousand GPS trackers, he's the most tracked criminal ever!" But of course every single one of those trackers was in sleep mode, and the only tracking bug was not part of the shipment. Sure, he was caught red-handed within half an hour while a police dispatcher watched a moving dot on a google map, but he was only being tracked by a single device, and we had no way to remote-active the other thousand.

Just an amusing related anecdote.

Set fire to the truck after you rob it to destroy any on-board evidence (like hard drives containing surveillance footage), only commit crimes in a wireless dead zone so no data is transmitted outside (or better yet, use a police stingray so you can do it anywhere), and put all cargo inside a faraday cage. Also I'm not sure there's any reason to believe those kind of security measures wouldn't already exist for a manned truck if the goods are expensive enough, so really it's the benefit of not having to deal with a person.

I'd much rather do all that than kill an armed truck driver.

Now that I actually am thinking about this more, I'd just get a sniper rifle and shoot the sensor units on the truck so it would be incapable of driving further (as opposed to throwing a mattress onto the road).

But again, I'm not in the business of robbing trucks. I'm just saying I'd rather rob a robot than a person. I doubt robberies are a big concern for driverless trucks.

Life isn't like a movie. Anyone who has the knowledge (of the vehicle and its surveillance systems, the schedule, the cargo onboard...) and the capability (skills and resources) to pull off what you're suggesting has plenty of lucrative legal opportunities. And any cargo load valuable enough to make a Hollywood-style heist actually worthwhile would be transported by more secure means.

It doesn't take that much skill to know that a certain stretch of road doesn't have cell signal and how to set fire to a truck. It takes even less skill when you know there's no human on the truck to stop you

It may or may not work, but I have no doubt that someone will try.

As soon as low orbit satellite internet services like Starlink are operational then you'll be out of luck, there won't be any more dead zones.

> Set fire to the truck after you rob it to destroy any on-board evidence

Trucks are big. They're also not particularly flammable. Hard drives are small and easy to fireproof.

> only commit crimes in a wireless dead zone so no data is transmitted outside

This put you pretty far out of your way, and you're eventually going to have to bring a truckload of goods back from the middle of Idaho or wherever. Plus, trucks can carry pretty big antenna.

> (or better yet, use a police stingray so you can do it anywhere)

This doesn't really work that way, and also this is an additional major crime. Plus, people are looking for these.

> and put all cargo inside a faraday cage

That's a big faraday cage. Plus, what happens when you have to take the cargo out of your truck-sized theft apparatus and fence it?

> I'd just get a sniper rifle and shoot the sensor units on the truck so it would be incapable of driving further

I doubt there's anyone on earth who could make that shot once, much less 5+ times in a few seconds. A truck is moving 60+ MPH and the target is a few inches across.

Also, every 10th truck probably has a guy with a shotgun in the back.

More to the point, I think if we're trying to secure the use case of "truck the same as it has been for 30 years gets a black box put in it" we're doing it wrong. After not too long you just flat out won't be able to get in, even if you do stop the truck from moving. You won't be able to drive it anywhere - what need is there for a cab on an auto-truck?

The problem is not doing the robbery itself. It's doing it again and again and again without being discovered/traced.

Serial killers have it easy: they can bury the body. You can't bury goods you need to fence. At some point there will be a pattern.

I'm not saying robbing trucks is easy. I'm saying if I were to choose between robbing a manned truck vs an unmanned truck, I would definitely rob the unmanned truck.

why? the driver of the manned truck is going to comply just as willingly. No way they are going to put their life on the line (unless they own the cargo). The unmanned truck will cameras and video uploaded, which is stronger evidence than eye witness testimony. Also police response will be faster on unmanned (they can't call the police during the hold up, you can take the persons cell phone). No way you can stop the distress call from unmanned unless you had an EMP or something.

> No way you can stop the distress call from unmanned unless you had an EMP or something.

Wouldn't be so certain. Jammers are illegal, but they aren't that hard to make.

If the truck is manned, it's robbery.

If the truck is unmanned, it's burglary (or theft, since a truck isn't a building).

They typically carry different penalties, and the difference is even more stark if you happen to have anything that could be considered a weapon on you when you do it.

People rob 18 wheelers all the time - there are particularly valuable loads from particularly interesting places.

Source: my father had to prevent this for film/cassette/DVD distribution, especially prerelease goods.

then robbery would still happen, but without endangering the life of a driver...

So I get it's a good thing?

> Then what? You have 20 palettes of dog chew toys or asian pears....? Some of the goods have serial/tracking numbers.

Your comment made me laugh. Very true. Besides the fact that there is no black market for 20 palettes of dog chew and asian pears, the other problem is that unless you successfully steal a few highly valuable items, you would need one or multiple trucks to carry what you've stolen from that semi. You probably don't want to use the semi you just stopped, because it could have multiple GPS trackers onboard.

If you knew that you were stopping a truck full of laptops, then maybe it would be worth it somehow, but as you pointed out, laptops have serial numbers... And then what, you load a few hundred laptops into a van, but your van is now "burned". It's been photographed and you need to dump it somewhere, further complicating your operation.

Then, I don't know, it seems to me like people just love to come up with imaginary reasons why self-driving cars/trucks can't work. They seem to fail to realize that, well, we can come up with even better safety measures. You could make your automated truck very hard to open. It doesn't need to have a lock that can be opened by a human with a physical key. It can have an electronic lock inside the door, shielded behind a 5mm thick steel plate, that's completely invisible from the outside.

You could also install a remote-controlled drone on top of the truck (value < $1000). The truck has cameras all around that record continuously, and as soon as the truck gets stopped (or even slightly before), it phones home. A remote operator sees everything the cameras saw. The thieves waste precious time opening and unloading the truck, and just when they're about to take off, the remote controlled drone starts up and follows them around for as long as its batteries will allow (~10-20 minutes), informing the authorities as to their position.

If robberies of automated trucks became rampant, I'm pretty sure we could come up with many ways to mitigate the problem. I mean, heck, we could even install pepper sprayers around the truck. But the robbers can just wear full-face masks, you say. Sure they can, but those masks aren't foolproof, and it's an additional piece of logistics they need to deal with.

You'd also have to dodge every other vehicle in the fleet, which are all technically drones.

> Then, I don't know, it seems to me like people just love to come up with imaginary reasons why self-driving cars/trucks can't work.

The way I like to put it (https://www.gwern.net/Complexity-vs-AI#technology-forecastin...) is: 'The critic asks “can I think of any reason this system might not work?” and stops as soon as they find one excuse, but the forecaster needs to ask, “can I think of any system like this which could ever work?” and keep going.'

I don’t think this is a fair characterization of adversarial attacks on AI. People on hacker news are simply forecasting the future of hacking. People aren’t really saying ‘it won’t work, forget about it!’ they are simply pointing out that historically technology and criminal enterprise have followed a similar pattern to the garter snake and the rough skinned newt[1]. The more sophisticated deterrents evolve, the more clever adversarial strategies become.

[1] https://www.atlasobscura.com/articles/how-a-deadly-camping-t...

> The more sophisticated deterrents evolve, the more clever adversarial strategies become.

And yet, not clever enough to be so successful as to stop automation, or economic development, or increasing global wealth, or...

> seems to me like people just love to come up with imaginary reasons why self-driving cars/trucks can't work

You do have a point. The core reality that self-driving is equivalent to full artificial intelligence and passing the Turing test should be sufficient I'd think.

I believe in incremental improvement. We can design vehicles that are increasingly more autonomous. They might never become completely autonomous (case in point, they will likely always need someone to come and repair them when damaged), but they can become autonomous enough to be useful. We're not far from the point where autonomous vehicles can become useful in specific scenarios, such as always following the same path on the same stretches of highway.

The thing I've always wondered about is: if governments can put street signs for humans on every road, why can't they do the same for autonomous vehicles? Sure, you could vandalize the signs and wreak havoc, but you can also remove a stop sign or hack a traffic light and do the same now. If a standards committee was formed to develop a spec for autonomous vehicle guides it seems like we could get to full autonomy far faster than waiting for AGI. Maybe you still have to drive on backcountry dirt roads, but wouldn't automating 90% of traffic be an enormous win for society?

Partial automation might be worse than no automation at all - if the human is forced to take over they'll be less prepared for it.

It’s relatively trivial to attach a shock baton or flash bang ordnance to the drone, which would incapacitate the thieves as well.

> people don't rob 18 wheelers

Haven't you seen the documentary, The Fast and the Furious (2001)?

You mean "Love, Death & Robots", episode "Blind Spot"?

Hijacking trucks (stealing trailers) is happening already with drivers and security guards. Remove the drivers and security guards and it will make it a lot easier.

"The Curious Case of the Disappearing Nuts" https://www.outsideonline.com/2186526/nut-job

"$370,000 worth of iPhone X devices were stolen from a UPS truck" https://www.theverge.com/2017/11/3/16601970/iphone-x-devices...

"As Freight on Trucks Becomes More Valuable, Thieves Get Creative in Their Attempts to Steal It" https://www.ttnews.com/articles/freight-trucks-becomes-more-...

Honestly though, that's win-win.

A heist where a human is involved has enhanced risk of injury or loss-of-life if the driver / guard decides to be a big hero or the criminal gets jumpy.

A world where vehicles are modestly easier to steal from (but audit the thefts) pushes risk from human lives towards insurance companies. Toss in a few honeypot shipments that are tailed by a handful of squad cars, and things are probably fine.

We don't have rampant highwayman theft because there are human drivers. We don't have it because most people don't want to do it.

It seems to me the heist risk is being overrated...

But if your logic is solid, the next solution would be for these self-driving trucks to carry a human or two who'll sit there and not do anything. So, literal human shields.

Like the joke about how to solve the self-driving car trolley problem to your benefit: make sure you're in a car with a few newborn babies.

I haven't heard that joke before and it's pretty good :)

But the human shields are still solving for the wrong problem: placing the value of a load of (usually) consumer goods over the value of a human life.

Feel like we already have ways to reduce this, having a lone trucker doesn’t magically make it all safer, if anything it should make it much riskier as there is a human life on the line.

If robberies start happening, insurance will adjust. Maybe a security driver will tail multiple vehicles headed to the same destination. Worrying about these cases seems helpful, but the industry will resolve these as the savings and improvements seem to make it a simple equation. This is all assuming the tech works ofcourse

> Hijacking trucks (stealing trailers) is happening already with drivers and security guards. Remove the drivers and security guards and it will make it a lot easier.

I'm not sure taking away bribe-able workers with keys to the truck will make it easier to hijack.

A person in the truck is the difference between a burglary and a robbery. One is a property crime, while the other is a violent crime. The law views them as very different.

People rob trucks all the time. It's actually a pretty big deal, especially in Canada for some reason.

See https://globalnews.ca/news/6419141/cargo-crime-greater-toron... and https://www.todaystrucking.com/canadian-cargo-theft-surpasse...

20 palettes of asian pears, (or any food item), is surprisingly easy to unload.

People absolutely rob 18 wheelers, and in creative ways too. There was a story in Sweden some time ago I found particularly crazy – robbers entering the back of a truck, while driving at about 50 mph. They climbed from the hood of a trailing car on to the truck, stole a bunch of stuff, and then climbed back out again. I wouldn’t have believed it if it weren’t for the fact that the affected shipping company (the national post service) rigged trucks with cameras to catch them in action. Mind boggling.

Also, not all valuable cargo are gps tracked smartphones and laptops, and not all situation mean having to stop and threaten a driver. For example: tree logs. They are often loaded onto trailers that are then left standing waiting to be picked up, or left standing part way to their destination because it’s driven by multiple drivers (usually happens when they have a long way to go.) It can take days before anyone even realizes the cargo is gone, because there’s a gap between drop off and pick up.

This kind of stuff is easy to steal, not especially hard to fence, and definitely not something your garden variety meth head does on a whim because of the logistics involved. This is enterprise level crime.

I’ve heard this is especially popular close to unmonitored border crossings (i.e. most EU borders) as it then makes the investigative work harder because jurisdiction.

Sure, gps tracking helps finding the stolen trailer – if it’s even being tracked – but by the time the police get there the goods are long gone and essentially untraceable. Nobody puts trackers in tree logs, and it can take days before anyone realizes it’s even gone because of the gap between drop off and pick up.

Just because it doesn’t make the news doesn’t mean it’s not happening.

Or it is their lucky day and full of N95 masks and hand sanitizer bottles!

Modern policing is trending towards recording everything and then backtracing who you are and where you came from once the crime is identified. So the truck would record the robbery/footage of the incident, this would go to a company specializing in transportation protection services which buys a lot of data from everyone. They will trace the instigators of the heist and send out a friendly swat team for meet and greet.

Isn't the bedsheet heist just as plausible with a human behind the wheel? There's a chance he won't see the people in the road or will choose to run them over, but if he does the decent thing then he's a sitting duck.

Well you'd have an eyewitness that you would need to attack, who is potentially armed. I imagine the solution would be comparable to whatever security protocols ATMs have v. bank tellers.

You are overrating how far truck drivers would go to resist a robbery. That's not their job, and the load is insured - even if a driver is there, they're not going to do anything to stop you.

I highly doubt any trucking company would risk the nightmare of having one of their employees fight back.

Companies have insurance for exactly this reason. And an eyewitness would be less reliable than just having a camera

> at which point the vehicle has no way to prevent a robbery

What a novel problem, truly unique to society. Goodness wonders how society protected stage coaches back in the 1800s.

Perhaps with deputized armed tax-payer funded civilians, tasked with investigation and law enforcement?

This is easy to prevent. First of all, it’d be crammed with cameras. If it’s truly unmanned, there wouldn’t be a steering wheel, so it would be impossible for a human to drive, even if a robber could somehow gain access to the locked cabin. The trailer can also be secure with heavy locks. I suppose you could brute force your way into the cabin, but you’d need another truck to load the goods onto. So what’s the plan, drive an empty semi down the highway in the middle of the night, somehow hijack and break into an unmanned one unnoticed, load up pallets of dog food or onions or whatever the thing might be transporting, quickly enough before cops show up? Good luck!

Yep, this is what doesn't make sense, if its so remote the cops cant get there quickly, its also remote enough the thieves cant get away quickly.

And a highway patrol type vehicle is coming in a bunch faster than lorries are getting away.

Occam's Razor here being to create some kind of automated firearm attached to the truck, the deadly superhuman speed and accuracy of which would act as a deterrent to freight thieves. This solution would have the extra benefit of adding some real suspense to the hijackings and robberies that do happen.

This is a serious comment and I can't think of any possible way in which this plan could backfire. There is nothing dystopian about automated trucks with AI-powered, high-precision guns filling America's highways 30 years in the future.

> Occam's Razor

I don't think that means what you think it means.

It's like Magnavolt, except works on people outside the vehicle.

only legal in 80s dystopian movies.

This is like pointing out how easy it is to rob a house or someone on the street. All you need to do is break a window (made of glass!) Or approach them alone at night with a gruff voice. Life hack! Why isn't everyone abusing this free money?!

Why would you rob a truck with omnidirectional and live internet cameras on every side of it, unload a trailer in haste, and try to make a getaway in another large truck, when you can rob a truck the old fashion way, by bribing the driver and have the entire trailer delivered and unloaded before anyone knows to look for it?

The risk vs reward are ridiculous to compare.

I think this discussion happened in too close proximity to the release of Fast and the Furious..

Two things I don't get in the current focus on self driving:

1. Urban navigation. Obviously the end game, but just exit to exit interstate navigation would be a much better near term goal and greatly benefit truck drivers along with anyone doing a road trip.

2. Unmanned. I would think it would be a long time before you would have unmanned trucks going down the highway. Seems much more obvious to have a driver / custodian in the truck for the foreseeable future until everyone is very comfortable with the reliability of the tech and its ability to cope with situations like you describe. Meanwhile for truck drivers to get around mandatory rest periods due to self driving assistance would be a huge win for the industry even if they are still manning the truck.

1) Short haul vs long haul segmentation already exists. self driving will disrupt long haul first. Long haul is endurance driving on long, flat interstates. Even if all you do is reduce the accident rate by 30%, you created tens of millions in value for the industry.

2)self driving trucks can pay for themselves just by having the truck on the road for more hours of the day. Laws prevent drivers from driving too much in 1 day. If the driver could legally sleep in the cabin for shifts, you can reduce overall costs a lot (without going full automation).

Highway is simpler but it's worse if it goes wrong.

Plus, highway driving is safer and cheaper than average so the bar is higher for self-driving cars/trucks.

I don't think there's any middle ground for driving semi-trucks on the freeway. It's either safe enough to drive without someone watching it or it's not. There's no point in having a person in there sometimes sleeping and sometimes safety-driving.

I expect you're right about keeping the driver around but napping. The robot can do the highway parts, and then wake up the human for the much more challenging parts, from highway exit to pulling up properly in the right loading bay.

And then I think the next step after that is to have local humans meet the trucks right off the highway, sort of like how pilot boats meet cargo ships and guide them into port. The truck cab stays empty for most of the run, and drivers get to sleep in their beds each night.

It's a mistake to think of this problem outside of the broader context of risk for shipping. Theft by external actors is just one of the many things that can go wrong when shipping goods. They can be late (leaving to spoilage in the case of perishable goods), they could be stolen by the driver, there could be an accident resulting in destruction of the goods. All of these risks get turned into pricing in the end, which is then combined with the actual cost of shipping the goods.

So the question is not whether there exist risks unique to autonomous trucking, the question is whether the cost in the end when taking all of these risks into account is lower.

If the control of the truck were to fail over to a remote-control system, like those used by the military's UAV's, that particular situation could be mitigated at least as well as it could by a driver on-site (can't hold a remote driver at gunpoint).

Nobody holds up railroads today because of the legal/law-enforcement system. If Amazon/Roadway's trucks had a bunch of HD video and lidar images of suspects, Law-enforcement would be off to a great start.

Furthermore, the loss to theft need only be less than the cost of maintaining a huge fleet of human drivers. The really big challenge, IMHO, is finding a fulfilling new career for the people that these systems might displace.

These vehicles will likely have plenty of surveillance surrounding them - not only for instances of heists, but for evidence to be used in insurance claims and police/fire investigations for collisions and other mishaps. Detracting from self-driving vehicles is puzzling to me, though. I would rather see the tech succeed and would like to see us get there as soon as possible. We should accept that we might have to adapt to them a little, and that our infrastructure may need to change in some ways to better accommodate AI. It's insane to me that we aren't laying sensors, reflectors, conductors, or whatever else would make self-driving vehicles safer and more effective in our roadways already.

I see these kinds of comments on HN so often and it's disappointing. It's almost like you're afraid to improve or event attempt to improve the world in any way because "Bad guys will do bad stuff".

Are you really so afraid of the world?

Do you think we should live in fear and improve nothing because of these "bad guys"?

If we modify something about society and then "bad guys" become more of a problem, they we'll just have to modify how we police and punish whatever that thing is too.

It's OK, you don't have to be afraid of the boogyman.

It doesn't have to prevent robberies, that's choosing the wrong battle. It just has to document the heck out of bad situations. Lock up, phone home, start filming, upload all the data you can, deploy a drone or two to do the same. We could probably think of some fun deterrent systems for cases where we're positive it's a theft, but really focusing the tools on gathering evidence is probably going to be a lot more useful.

Aside: is it robbery if you're attacking a passengerless automated truck?

You don't need blankets or elaborate decoys.

You just need to stand in front of the truck and don't go away. You should start far enough away to give the truck enough road to stop. On a wide highway you will need more people to block enough lanes.

Definitely a driverless cargo truck should not hit a human.

Works with human drivers, too. So if it was worth it, such heists would happen often enough for the public to be aware somehow.

Probably the risk / reward ratio is not good here. Transporting the loot secretly must be especially problematic.

Yes, after we eliminate all the trucking jobs in the county which are one of the last remaining decent jobs for unprivileged people, then the robberies will really take off. Better come up with a log(n) solution to those dirty poor people with bedsheets before prioritizing anything else, because there will surely be robberies.

Maybe instead of intentionally creating a nightmarish hellscape so we can fix the robbery problem we just take a step back for a minute? What are we actually building? For whom?

If I wanted to bring this discussion even further into science-fiction territory, I'd bring up the idea that the AI truck needs to be armed so that it can defend itself.

The truck will be collecting storing and probably uploading a huge amount of evidence from many kinds of sensors during such a heist, so good luck getting away with that. Would make for a good Ocean's movie.

It’s a good point. However, I doubt anyone will be transporting high value items in autonomous vehicles for quite a while, as insurance will be expensive initially.

All evidence points to instrumented vehicles being less vulnerable. Vide all the YouTube videos of vandals and thieves caught on "sentry mode" videos.

Automated video collection devices collect video, so they produce more video than not using them. That doesn't mean they create more security.

You could have one human teleoperator monitoring 10 trucks for corner cases like that. In your example the human could take over and drive over the bed sheet

And what exactly would a human driver do in this case? More than likely the company policy is already "do nothing, insurance will cover it".

Trucks are usually stolen and not unloaded in the middle of the highway. The theft is then alerted to law enforcement including weigh stations.

A remote operator immediately connected with the right heuristics could easily prevent a robbery

Yeah, the cops put a sign near my work on the highway saying about cargo robberies.

That is not an engineering problem, its a social/political problem.

Stealing from a regular truck with a lone driver risks becoming a robbery, because there is likely a human around. That alone deters criminals.

Self-driving trucks have no such probabilistic deterrent.

Predator drones?

Reply on news.ycombinator.com