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.
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.
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 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.
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.
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?
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).
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.
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.
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).
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.
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.
> 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.
> 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.
> 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?
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.
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.
> 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.
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
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.
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.
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
> 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?
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.
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.
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.
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.
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!
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.
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).
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.
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.
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 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.
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
If Waymo is successful, there are potentially massive impacts to the population of the US. Trucking is a major industry.
The Luddites weren't mad about the advancement of technology, they didn't destroy looms out of religious or anti-technological ideals. Many of them were skilled loom operators. They were squeezed out of their jobs by the race to the bottom, and they were angry that there wasn't any consideration for the paradigm shift they were experiencing.
If Waymo turns trucking into an automated business (or even just reduces the number of humans needed to haul cargo), we could be in for a similar situation. I wonder what Waymo is doing to help support the existing drivers, regulating how we should care for the many millions of humans this could impact, etc.
I believe automated vehicles will someday be safer and more desirable, but we need to ensure that they aren't also the villains in the disenfranchisement of millions of people.
This is the problem of modern democracies in a nutshell. We saw the same thing with free trade. International free trade and movement of capital is claimed to increase the overall wealth of a nation whilst massively negatively impacting a small portion of the population (outsourcing clothing manufacturing gives everyone cheaper clothes but makes clothing factory workers in 1st world countries redundant). The huge corporations that profit the most from these new economies of scale and international competition then lobby to reduce social safety nets and standards so the exact people who were negatively impacted by these changes receive no help to get themselves back on their feet.
If it really is true that these changes massively increase the general wealth then it would seem trivial for governments to step in, take a small cut and put it towards retraining, re-employing and retiring these workers. Yet it very rarely happens.
A big part of government duty in capitalism (and often forgotten) is to keep a level playing field. That is what capitalism thrives on, competition but restricted to stop powerful groups exploiting others.
And I think this is important if we want to continue in free trade. We need to do things like have better requirements for pollution, global worker treatment rules and all sort of things so countries aren't competing on who can abuse people/enviroment the most, but on innovation and skill set. Further I think we need some global consensus on minimum tax rates. These aren't traditionally in the WTO as far as I know but feel they need to be working in in the future to create something closer to a level playing field to be part of any trading block.
Capitalism goes as far as making sure there are at least two competitors in a space, and with all of the companies going at self-driving trucks there won't be a problem there. Other than that, the goal of capitalism is to give consumers [big box stores] the best possible product [Delivery trucks] at the lowest possible prices [not having to pay a human 50k/yr].
> Capitalism goes as far as making sure there are at least two competitors in a space
There is no absolute truth here but I would say generally that is not true and it is encouraged to be more than a few companies, but is increasing allowed today.
Typically Oligopolies are frowned upon but there is often 1) Industry categories that seem to be how they are and 2) With increasing business influence in politics this is less regulated today than previous.
I hope the government does, but I also think it's in Waymo's best interest. We're not talking about a closet industry here, trucking employees ~8 million Americans, and probably supports many, many more than that.
You've seen the fuss over coal miners in the States as of late? There are ~50,000 coal miners. At its peak, coal mining was less than a million miners. The potential civil unrest from disrupting trucking could be a big deal. From insolvency of truckers who bought their own (now obsolete) equipment to demagoguery of politicians trying to soothe an angry group of people.
I doubt it'll be a like a light switch where one day there were no Waymo trucks, and the next day there are no conventional trucks.
It will likely be much closer to how Uber disrupted the Taxi Medallion market. You don't hear many people saying, "But think of the poor Taxi Medallion owners." But -- actually -- that was worse. It's literally a piece of paper.
If you buy a truck today, and self-driving trucks are a thing tomorrow -- you still have truck that's likely still holds most of its value.
Part of the reason few people said "think of the poor medallion owners" is because the whole medallion system was a deeply corrupt racket run by organized crime.
Coal miners didn't just light switch off either. And I definitely heard people saying, "Think of the taxi drivers" (which I know is technically different from the medallion owners).
These concerns are always overblown. The switch to autonomous vehicles isn't happening overnight. It'll be a generational change, and the current drivers will simply retire. If there's a young kid today who is dreaming of growing up and becoming a truck driver - yeah he's in trouble.
Depends on how fast it happens. Long-haul trucking is the most easily automated driving, and it has the strongest financial incentive. The lifespan of a semi truck is ~15 years, while the career of a trucker will be 40 years. Even if 2/3rds of truckers retire before automation hurts them, that still would leave us with more than a million long-term unemployed.
Only if they automate it. But they don't have to. They could have local drivers rendezvous with long-haul trucks as they get close to destination cities. That would still let them cut their labor costs by 90%.
The problem is no one is saying not to get in to these industries. In fact quite often there are posts on forums and even on HN stating how you should become a truck driver or coal miner because it pays well and doesn't require education.
>If Waymo is successful, there are potentially massive impacts to the population of the US. Trucking is a major industry.
Certainly, but we don't know if it will be positive or negative. Some studies[0] show increases in human truck driver employment with long-haul being fully automated (plus the work seems nicer).
The source for that article has been deleted. Additionally, the study was performed by Uber, who has an incentive to suggest that automating vehicles is a good thing.
If you have other sources, I'd be happy to take a look.
I hadn't heard that description of the mentality of Luddites before. Thanks for bringing that up.
It may be different for trucking for a few reasons
1) apparently there is a huge shortage of truckers in the US currently
2) autonomous driving may be great for highway today, but seems to still be a bit away from truly autonomous for city environments. This could lead to a hub & spoke structure which keeps truckers employed, but has them taking over for more local routes.
3) As I understand it, a mechanical weave was always significantly faster than a person, and cheaper to boot. An autonomous truck doesn't get to drive faster (yet), and doesn't burn less fuel. It's efficiency comes only from replacing the person behind the wheel. If the person behind the wheel owns the truck (which is common) it may make their work easier.
4) how many people these days are wanting to become truckers? As truckers retire, and we seem to be requiring more and more shipping (an entirely different issue), could this be a managed transition over 20 years?
definitely a good point, I was considering more of a hybrid approach where a driver exists but some of the driving is done autonomously. We're close to that now.
Also, trucks still need to refuel, so breaks are still required.
Did car companies have to support blacksmiths, especially the ones skilled at shoeing horses? Airlines aren’t forced to subsidise railways. Innovators aren’t obligated to support the industries that they supplant.
But consider that there are probably 10 million+ Americans who rely on trucking/trucker incomes. If a significant portion of them are put out of work, it's a recipe for massive civil unrest. Not to mention the impact to our stretched social safety nets.
Waymo doesn't have to do anything here, sure, but if they don't then I'd expect we see Luddite 2.0.
I personally think that innovation should be done ethically, that you should proceed forward, but don't be blind to the impact you could have.
Trucking will go the same way - it is far more efficient (and allows you to grow faster), therefore market pressure will drive every trucking company to autonomous.
Yes. This is exactly my concern. Long term, I expect we're all better off, but we cannot just look at the present and the long term, we need to understand how disruptive it will be in the interim and make sure we're supporting the humans that are currently doing the work.
> "If Waymo turns trucking into an automated business (or even just reduces the number of humans needed to haul cargo), we could be in for a similar situation."
Sure.
Tuning in to last week's news - "Mass layoffs reported after Starsky Robotics fails to find buyer, investors" [0]. Starsky had a demonstrable path to "reduce the number of humans needed to haul cargo". What gives?
I'm not sure if drivers will be completely eradicated, though. The vehicle may still have an operator for non-highway situations, and given there's already a shortage in truck drivers, perhaps the career will slowly ramp down as automation slowly ramps up. I used to think SDC and automation would be fast and furious, but the opposite has occurred so far.
Sure. In the Luddite's case the number of workers wasn't dropped to zero either, but it did mean they could hire fewer, less qualified workers. I suspect the qualifications and pay for trucking minders will be different than current truck drivers.
Maybe it'll be slow enough that we naturally see less demand gradually. But we should still have a plan for those folks in case it's rapid.
“what Waymo is doing to help support the existing drivers, ”
Waymo should be doing nothing. Companies should work on improving technology. The problem is to distribute the gains. That’s what social safety nets are for. And progressive taxation should help to distribute gains from technological progress widely instead instead of only to a few capitalists. I am aware that this may be an unpopular opinion.
I mean, totally. Gimme that progressive tax structure and strong social safety net any day of the year. But that sounds like a harder problem to solve in the US than autonomous driving.
As a bike/ped advocate, I believe self-driving cars will ultimately be safer than humans. Yesterday I was nearly hit crossing street twice: once by an inattentive driver and once by an intentionally reckless speeder.
Oh the other hand, my brother has done a fair amount of truck driving and I'm aware it's a common job in the US.
Maybe now that Bloomberg has dropped out he get together with Yang and use his billions to start providing a basic income to displaced truck drivers?
I don't know if skewed over imprecise. There's all of the ancillary industries that support OTR trucking affected as well - that'll bring the total numbers back up.
It seems like Yang has written something nice on Twitter about every candidate who dropped out? He seems to like people and praise doesn't cost anything.
This is like when people started calling Glenn Greenwald an "alt-right supporter" because he said he admired the rise and efficacy of Brietbart's operations some time during or after the previous election (can't remember when exactly, it was after they skyrocketed from fringe site to a major news operation).
Social media is turning sections of the public back a few centuries where mere a positive reference is guilt-by-association and the tribalistic off-with-the-head mentality kicks in.
It's quite nasty and anti-intellectual, and the growing prevalence and toleration of it for political ends honestly scares me a bit.
Yep, as a biker I now intentionally try to look at the driver through the windshield to see if their eyes are down on a cell phone. They are scarily more frequent than you would think.
Or just to check to see if they're looking in your direction, so they're at least likely to have seen you. Amazing how often people watch a car to (say) their right, and once it clears, they go through the intersection without checking again to their left...!
As a driver, every once in awhile I glance at the faces of other drivers on the road.
Sometimes I get the feeling that more than 50% of road-time in the US is spent looking at a phone.
It's enough to make me wonder if "x% of accidents involved texting and driving" is even a convincing metric. If god damned everybody is texting and driving all the god damned time, of course lots of accidents involve texting and driving.
> use his billions to start providing a basic income to displaced truck drivers?
Are there no other viable options to "truck drivers being replaced by automation" than basic income? I've read a lot of pushback about basic income online.
I don't know the answer, but one problem is that truck driving is already often a plan "B" (or "C", or ...), in that it provided a practical home for workers displaced from other industries. Often one they didn't have to move to get.
This has difficult implications if you are looking for things these folks can do instead of trucking.
You're talking about a demographic that is largely old, in poor health, and with few other skills. They're not going to see a huge RoI from training/education, and they're not attractive to employers. At the moment, their value proposition is that they're doing an unpleasant, unhealthy job that not many other people want to; it's hard to see how that would translate into good employment opportunities in other sectors, or even if we should want it to.
yes, UBI's first- and second-order effects are designed to corral the citizenry into more control and dependency on the elite, rather than the stated intent of providing security and freedom.
besides, it's just as straightforward to offer every citizen a part-time job with the government that's worth $1000/mo and avoid most of its unintended consequences.
i would have supported yang more had he not staked his whole campaign on UBI, as he'd otherwise made some sensible arguments.
Sometimes I wonder if inattentiveness (i.e. breadth of thought) is perhaps a fundamental attribute of generalised intelligence, and having a laser focus on the task is the wrong way to get there.
With how much money everyone is spending, wouldn't it just be cheaper to build a private cross-country interstate where only AI drivers are allowed. Let humans drive the trucks to this interstate then let the AI take over. It seems like almost all the edge cases would be solved if you just got rid of other human drivers and let the vehicles network with one another.
> Total cost of interstate highway system: $128.9 billion [1].
That's in 1991. So with inflation, it would be $244 billion today. Also, I suspect the $128.9 billion figure itself was not inflation adjusted.
> Total invested in self driving technology as of 2017: $80 billion [2].
Wow, that's a lot more than I expected. But you have to keep in mind that once self driving is "solved", it will work on all highways around the world. If you think about how much it cost to build all the highways around the world, it's way past the trillion dollar mark.
you could even put down tracks for the cars to go on so they could reach higher speeds, and chain them together for efficiency. almost like some kind of train!
The problem is that the whole point of trucks is that they can go almost anywhere. Replace that with dedicated roads and you might have cheaper unit economics or whatever but you still have a rail-like distribution network.
Is there room to build a second interstate system on top of the one that already exists? Seems like the cost could be a lot higher now than it was when there was no development.
There's a rideshare company called Via. Is this intentionally designed to take away their SEO?? I'd be surprised if there isn't a trademark challenge for this, they're a competitor with Waymo.
That's not how trademark law works though. Just because apples are a common, 20 million year old fruit doesn't mean Apple can't trademark the brand Apple in relation to computers and personal electronics.
Trademarks matter in context a lot. An Apple branded computer is a lot stronger than an Apple branded apple.
They choose a brand name bordering on descriptive. If it's descriptive, ie "Great Computer Co.", it's pretty weak. It'll be tough to own "Via" for anything in transportation.
Lumber hauling off road may actually be one of the easier applications because the road is usually dedicated to trucking. Logging trucks already assume they can blast along at full speed because if you use a logging road, you’re supposed to radio your position via CB to ensure the trucks know where you are and don’t hit you.
Sigh. Forest roads are absolutely not dedicated to logging, they are frequently used by hikers, backpackers, cyclists, and equestrians, and just people driving to a picnic spot. At least in America working forests are also public lands.
This is true, but nonetheless you’re supposed to carry a CB. On Canadian logging roads, this is not something they take lightly - at least on the signage!
I believe it can drive during winter, just not in extreme condition (snowstorm). Of course, it can simply wait until the storm passes and continue. And honestly, many of the cases where Waymo can't safely drive, realistically most humans shouldn't be driving either but we still do because we're stubborn like that.
OTOH, if there's a 24h stretch of good condition, it can drive the whole time.
For long haul you'd still want someone riding in the cab who can do basic maint I think, plus off loading the truck a lot of the times is the duty of the driver. Some companies advertise 'never tarp again!' because tarping your load on flatbeds is the driver's responsibility. Also for in-transit maint airlines need repair, trailer power needs to be checked especially for reefers. I'm not sure about otr truckers but for ltl the driver is offloading trailer.
I can see the training and responsibilities shift as things become more automated, but for what a driver costs, I can't see them getting rid of an incab person for otr trips which leave the city. The response time and costs for fixing minor maint like a broken air line would be crazy. Also someone gotta fill those tanks up.
Why would you have the driver ride along for the long-haul, rather than have the truck drive autonomously between cities and then pick up a driver once it gets to the destination metro area for last-mile/unloading/maintenance?
The bulk of the time (= labor costs) from trucking is on the interstates, and as an added perk, this would allow truckers to work 9-5 and stay close to their families rather than being on the road all the time.
Using a mix of automated and human drivers (aka, the AI is one half of a team) seems like a great way to go. Consider that there is very little freight that goes longer than 500 mi without either being on a team or on the rail.
Professional driving comes with hours limitations. In most industrialised countries you can't work more than so-and-so many hours driving without so-and-so many hours of rest (I don't know what the US rules are but I know they have them). But as just a passenger, aboard the truck but not driving it, you aren't using up those precious hours.
I can see this making sense where the driver also has some skilled role at both ends, e.g. supervising loading and unloading of a particular type of cargo. Otherwise it may be just as sensible to build giant truck parking lots at key locations on big highways and have a human board trucks when they get to the closest lot to the final destination. In that model all truck drivers would go home every day, which means it's a less well paid job but also one that's more compatible with having a life outside work - long distance trucking would become something trucks do but drivers don't.
>But as just a passenger, aboard the truck but not driving it, you aren't using up those precious hours.
This is inaccurate. If you are sitting in the passenger seat you are classified as "On-Duty Not Driving"
You are only 'resting' (aka "Off-Duty") when you are in the bunk.
This however is frequently violated and widely ignored, as there is no enforcement mechanism, when I teamed, I'd do my 8 hours of sleep, then come up at sit with my co-driver and talk with him about the day ahead as he drove until it was time to stop
From the placement of the LIDARs, I wonder how they solved the problem of being able to look in the truck's blind spot. Or do they just attach a camera on the back of the container every time they switch containers??
> We’re currently testing our fleet of trucks in California and Arizona, and we’ll soon expand to Texas and New Mexico in the near future.
"soon <thing> in the near future" is a weird structure. It's repeating itself, which either comes across as poorly constructed or _really_ emphasizing. I doubt it is the latter....but if it were, I'd expect waymo to announce Texas/NM support within a few weeks. (Which begs the question of "why publish this blog now, instead of in 1-3 weeks when Texas/NM support launches?")
Are these semis truly driverless or does the testing that Waymo is conducting still require staff to monitor the drive? I want to parse through the hype here.
Fully automated, point-to-point, driverless semis seem far off, as Hotz of Comma.ai was pointing out. Am I wrong on this assumption?
> Are these semis truly driverless or does the testing that Waymo is conducting still require staff to monitor the drive?
I would assume they're planning to follow a similar trajectory to their passenger vehicles.
There are almost certainly human monitors on the trucks that are running right now, but the long term goal is to be driverless.
> Fully automated, point-to-point, driverless semis seem far off, as Hotz of Comma.ai was pointing out. Am I wrong on this assumption?
I think fully automated arbitrary point-to-point anything is a long way off at best and potentially impossible. Being able to handle any situation, including those that often give human drivers trouble, seems like it very likely could be a "hard AI" problem to me
I think that shrinking the problem space to a fixed route makes things a lot simpler, especially for deliveries that aren't extremely time critical where the vehicle could safely pull off if things got questionable. Think about something like a load of cars being shipped from Detroit to Denver.
Even if the autonomous trucks are limited to interstate-class highways, it could work like Turnpike Triples do here in Ohio where the autonomous truck could handle the interstate part and then drop the trailer(s) in a transfer lot located right at the exit. Human drivers could then pick them up from there for the local portion of the haul. That seems like it'd be a win-win, the companies get their loads able to drive any distance with no time limits while the drivers get a nice local loop where they're home every night.
anybody knows - is that new slim sensor package a major upgrade to the set/quality of sensors or just a re-packaging? They've only recently appeared in MV, so i haven't been able to get a good look yet :) It is interesting that it appeared on the wave of doubling of the Waymo staff in 1 year plus major outside investment and all those new announcements. It makes one wonder whether they've got major improvements or are only gearing toward it and/or just preparing to expand the operations for one final push across the finish line.
Arizona basically threw their citizens to the wolves with a 2015 self-driving car law that allowed anything and everything to go (you wouldn't even have to register your cars as self-driving for a period). This attracted tons of firms to test their cars in a lax regulatory environment but since the famous Uber death and a number of self-driving accidents, they have tightened the rules somewhat. I suspect they're still looser than average and there's some inertia from everyone setting up shop there originally.
From the original press release:
> "Part of what makes Arizona an ideal place for Uber and other companies to test autonomous vehicle technology is that there are no special permits or licensing required," the ADOT release said. "In Arizona, autonomous vehicles have the same registration requirements as any other vehicle, and nothing in state law prevents testing autonomous vehicles."
Arizona also allows people under 30 to drive, and every age group under 30 has a much higher accident rate than 80+. People under the age of 18 have quadruple the accident rate of people aged 80. The reason people incorrectly believe that octogenarians are bad drivers is they have a tendency to die in even minor accidents.
I think folks think 80+ drivers are bad, is they drive conservatively and carefully. Its annoying to 20-something drivers who want to race around and take chances.
See, the thing is, Waymo controls the data, and certainly knows how good their driving is. If it was clearly better than the average octogenarian driving, wouldn't they be letting us all know that? Wouldn't they be announcing that and writing up white papers and giving presentations at conferences about this? They need to start building the safety case, and it is entirely to their benefit to release this data, if it truly is more reliable than even just the average octogenarian. The fact that they don't, and so far as I can tell only release the minimum data that they are legally obligated to (and marketing-speak like https://waymo.com/safety/ is not the same thing at all), leaves me with an inference that the data is not nearly that rosy.
There's a difference between something being safe and being capable. I can make a trivially safe self driving car that just sits there and does nothing, completely safe but not capable :) My point being that Waymo could have zero colissions (which I believe is not true) but still not be as capable a driver as someone for which it is unsafe to be driving, but is reasonably capable of getting around.
Both lack of regulation as well as weather, road condition, and road layout. Most of the Phoenix metro area is a grid, fairly modern as streets are constantly redone as we've grown so new-ish signs that are largely unobstructed by trees, etcetc.
As a motorcyclist with a 45 minute commute, I'm on the cautious side when it comes to self-driving tech on the roads, but I have to say that anecdotally it's been fine so far. I'd be willing to bet I see ~10 Waymo vehicles a day. I've had plenty of issues with normal cars/drivers, have never had nor seen an issue with Waymo.
Uber was testing there because the local city government was easy to work with initially. After their fatal incident, the city became more restrictive, but Google/Waymo had already also moved in and started testing there, and because they didn't have any issues, I presume the city continued to work with them.
I've been wondering for years why nobody mentions this. I saw them on the road in Sunnyvale in 2018. Maybe people assume they are just regular trucks used to transport Waymo cars or something.
You see HN comments all the time telling startups to put up a page before their product is ready. It's super common advice. Here are a few offhand reasons I can suggest it's not a waste even if the product is not for sale.
Marketing. Lead times for delivery contracts are probably long. Expanding your marketing, making your target audience aware of the product ahead of time, is useful. It makes future sales cycles easier, as your potential customers may already know about your new product.
Hype. Self-driving cars and trucks already have lots of hype, but not usually for the big 18wheeler delivery style. Waymo may need to increase not just awareness but also excitement.
Testing for product-market fit. While a purchaser won't make a decision based on this page alone, Waymo can make some decisions and learn some things based on people's reactions to this.
I could keep going, there are thousands of reasons that making a page for a product or service not yet for sale can be a good idea and not a waste.
Everyone (Pony.ai started trucking as well) is converging to trucking as the canonical use cases. Let's see how this affect the existing self-driving firms.
In hindsight, trucking seems on the more obvious usecase for autonomous vehicles rather than daily commutes or ridesharing.
1. Most time spent and miles driven are on freeways rather than city streets. Last mile can still be done via human drivers meeting at an off-freeway ramp hub.
2. The cost and availability of labor is a significant damper on trucking. Remember during the oil boom in Alberta, Canada, how truckers were making six-figures and higher due to the high demand.
3. Humans get tired, can't drive for more than 8 hours due to legal restrictions and safety concerns. These are problem that machines can solve.
4. (Mostly) don't have to deal with pedestrians, bicycles, bad weather and other complexities. If there's a snowstorm, trucks can just do what they do now, wait it out at a truck stop.
In foresight too. This has been part of the conversation since the very beginning; pointing out that the most likely initial targets for "real" commercial use at scale was long haul trucking, with people for either end. Usually predicated by heavy machinery in places like mines, where variants are already done.
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.