Hacker News new | comments | ask | show | jobs | submit login
The driverless truck is coming, and it’s going to automate millions of jobs (techcrunch.com)
447 points by rottencupcakes on Apr 26, 2016 | hide | past | web | favorite | 455 comments

My father is a truck driver. He made his living driving vehicles and doing heavy lifting and menial work. Truck driving is just the latest evolution in his "career" and I'm ever grateful to him for battering his body in various ways to ensure I had every opportunity available to myself.

We talk a lot about his job. It's not fun. He works long hours, a lot of it is mundane driving, and he sits idle a lot of the time.

I think he would be the first to agree that automation of his job is inevitable and likely necessary, given the dangers of truck driving. Hurling a multi-ton vehicle to and fro is a dangerous task at the best of times.

But therein lies the rub. We are just now reaching the point where we as a society are getting comfortable with automated cars. I think people will be less comfortable with the idea that the truck next to them has no human in it, and could experience some kind of glitch with catastrophic consequences. HN readers will understand that these cars are still a ways away, since the last 10% of the work required for true automation will take a lot longer to develop.

Trucks are way more complicated to drive. Once they're up to cruising speed, they're easy. But it's everything leading up to that point which is hard. Dealing with gear changes, airbrakes, load shifting, other vehicles that will inevitably cut you off, and much more.

And then when you get to your destination, manoeuvring a 50' trailer is no easy task, even for a trained driver like my dad.

It will happen, and it should happen. But it won't be easy, nor soon. Of course, I'd be happy to be wrong, and I'm sure my dad would be too.

> Trucks are way more complicated to drive. Once they're up to cruising speed, they're easy. But it's everything leading up to that point which is hard. Dealing with gear changes, airbrakes, load shifting, other vehicles that will inevitably cut you off, and much more.

The mechanical aspects of controlling a large truck and accounting for the variables is probably the easiest part of the problem.

Things involving humans (vehicles cutting you off, pedestrians, etc.) are part of the general class of problems that I'm assuming all automated vehicles are working on and are much harder.

Yes. Truckers have to deal with a sort of standard "road safety dilemma" all the time, which can be boiled down to this:

A trucker is driving a 40 ton truck at 65mph. A passenger car with a family of four cuts in front of it, and the distance is closing fast.

A passenger car takes roughly 315 feet to come to a complete stop. A fully loaded truck takes 525 feet. The trucker now has to make a decision.

A) Do I slam on the brakes and keep going straight, and completely obliterate the car in front of me and the family inside it?

B) Do I cut off to the side, running over the rail, flipping the truck, killing myself, and possibly any other vehicles or pedestrians around, behind, or coming in the opposite direction of me?

The robot trucker could of course decide on B), but there are unknown variables about the extent of the resulting damage. This risk increases with merging between lanes or on/off ramps both by passenger vehicles, the truck itself, and motorcycles.

A dedicated lane would make this fantastically safer, but who will pay for it?

>A passenger car takes roughly 315 feet to come to a complete stop. A fully loaded truck takes 525 feet.

How does this change if truck design and "per-axel" fee incentives were rationalized[1][2] to instead maximize the number of axles? Doubling the number of axles reduces road damage by 94%. Currently semi trucks account for 6% of vehicle miles driven[3] and 99% of all traffic damage to roads.

But would this measure also improve braking distance and safety?

[1] http://www.nytimes.com/1989/07/26/business/economic-scene-be... (erroneously states that damage scales as the cube of axle weight; it scales as the fourth power or even the sixth(!) power on weak roads)

[2] https://truecostblog.com/2009/06/02/the-hidden-trucking-indu...

[3] http://www.princeton.edu/~alaink/Orf467F13/FreightFacts&Figu...

Braking distance is a factor of speed, weight, brakes, tires and road surface. A 40 ton truck can only stop so fast with rubber tires on typical brakes on a typical highway road surface.

Race cars are designed to stop faster than passenger cars. Their brakes are huge, they're designed with exotic components to transfer heat to air faster, designed to take higher heat for longer, the cars are designed to flow air over the brakes, and they cost much more and don't last as long, and have advanced computerized systems to optimize brake utilization. They also have pretty big tires (usually for power transfer, not braking). Their tires are also grippier on a given surface at a higher tire temperature, so they cost more and don't last as long, and work best on dry flat surfaces. They strip all the weight out of the car to decrease the amount of force working against the tires and brakes. Even the suspension and shell is designed to increase tire, and thus brake, traction. They stop really goddamn fast - 93 feet for a Ferrari F430 Scuderia, from 60mph.

Trucks are designed to carry 40 tons up and down hills, survive tire blow-outs, brake system failures, and more miles than 1000 F430s will ever see, on all kinds of road surfaces and temperatures. It is possible to improve the brake time, but it would probably cost so much you might as well ship your produce in a Ferrari. Halve the weight they carry to reduce braking distance and now we have to have twice the trucks to carry the same loads, but you still have a vehicle 10x heavier than other vehicles with dynamics that simply can't move around like the lighter ones. More accidents would be guaranteed.

> Race cars are designed to stop faster than passenger cars. Their brakes are huge, they're designed with exotic components to transfer heat to air faster, designed to take higher heat for longer, the cars are designed to flow air over the brakes, and they cost much more and don't last as long, and have advanced computerized systems to optimize brake utilization.

A lot of that isn't to brake faster but to keep the brakes from overheating from constant high use during a race that can last a couple hours or more. For normal emergency use you can drop a lot of that because they only need to be used extremely hard rarely and they'll have time to cool.

In other words, look at aircraft, not at racecars. What you have described is practically the only mode of operation for wheelbrakes there ("brakes catch fire during emergency braking? No biggie, just keep an eye on them while they cool.").

Brake heating is not a real issue with emergency braking, only with frequent braking which race cars do. That's why "upgrading" to "big brake" systems on passenger cars is usually a waste of money and reduces performance: unless you're driving on a track where you have to brake a lot, and the cumulative braking keeps the brakes hot, you're not helping. Even worse, the brake pad compounds used for race cars are different, and are designed to maintain performance even when very hot, whereas passenger car brakes have to have great performance when cold.

The bottom line is: a single panic stop on the highway isn't limited by your brakes on a passenger car, it's limited by your tires. The more friction your tires have with the road, the faster you'll slow down.

So yes, if you want a 40 ton truck to stop faster, it's completely doable: just increase the amount of rubber in contact with the road. Adding more wheels will do that quickly. However, more tires means more rolling friction which means less fuel economy and fuel economy is critically important to long-haul trucking. Also, adding more axles causes maneuverability problems in the city. Notice that on standard trailers, the rear axle set can be moved forward and back. They set them in the rearward position for long hauls, and in the forward position for shorter hauls.

If there's no driver in the equation (no monthly salary to pay), I guess the the incentive will shift towards having more, smaller trucks.

Not necessarily. More, smaller trucks will mean your transport costs will go up because you get less economy of scale. Eliminating the driver helps mitigate that of course, but going to smaller trucks then eliminates the savings you got by eliminating the driver. Having more, smaller trucks would be more versatile however, if you don't need to transport full truckloads of cargo between certain points.

Also, drivers as I understand it aren't paid monthly salaries, they're paid by the mile.

The article mentions driverless trucks would drive slower, at 45mph, wouldn't this make a key difference in this scenario, eg a much shorter breaking distance. Also, a driverless vehicle has one option a manned vehicle doesn't - self destruction to save life, eg drive off the side of the road and crash (assuming it could determine that no one was there)

That'd be a disaster on single-lane roads. Even with human trucks and their help (signalling if road is clear, driving on the shoulder when possible etc), it sucks to overtake them. If trucks were limited to 70km/h, there'd be much more need to overtake them. Which would either cause either traffic congestion or more drivers doing reckless things. Let alone that non-driverless trucks/buses/campers/etc would be stuck behind them forever.

I imagine they would start out only on multi-lane, limited-access highways. The small number of entry and exit points would make it much easier to program the truck. Such roads carry the bulk of long-distance truck traffic anyway.

Once you get away from highly populated and developed areas, there're lots of single-lane roads. Let alone that single-lane roads are used as backups when inevitable accident happens on multilane. Of course, there's not as much traffic in such areas as in those with multi-lane highways.

Well, the current tests with self-driving trucks are almost exclusively by european carmakers on German roads – where 2 to 3 lanes are common about everywhere, and trucks are limited to 80km/h or 100km/h anyway.

>. Also, a driverless vehicle has one option a manned vehicle doesn't - self destruction to save life,

Don't they?


It's difficult enough for driverless cars to know where the lines on the road are. I don't think we can expect in the near future to be able to not only detect pedestrians, but to have an idea what series of events may happen during an avoidance maneuver. There's a lot of things to go wrong.

However, stopping distance in real-world scenarios is wayyyy farther than in practice runs, because even a second faster braking by the driver can save a hundred feet. So robotic braking (that we have today in passenger cars) could improve the distance a lot. But...

Trucks are still 10-20x heavier than passenger cars, and there are practical upper bounds to the size of their brakes and the g's they can pull. If a passenger car normally stops in 388 feet at 65mph (not uncommon) and it would stop at 196 feet at 45mph, a truck at 45mph would stop in 265 feet. So yes, the truck could in theory stop faster than the passenger car.

But passenger cars don't just slam on their brakes in front of trucks (often). They also just drift right the heck into the truck. Perhaps mechanisms could be developed to safely merge away, and they'll probably be developed in passenger cars first. But there's no denying that a dedicated roadway would not only save repair costs for passenger-vehicle-only roads, it would remove the risk of colliding with passenger vehicles no matter whose fault it was.

There's a real possibility that once driverless trucks become widespread, human behavior will change as well. A very aggressive driver may take advantage of the fact that an automated truck will brake for them, and start routinely cutting off trucks, whereas today they'd be a little more respectful. Even pedestrians could start crossing a highway where they shouldn't and know they could get away with it.

The next step in this progression is that trucks report cars being purposefully a danger to them, with video evidence.

A robot trucker will never do (B). The car put itself there and can presumably get itself out again. More importantly, swerving will lead your truck to jackknife and roll in unpredictable ways and cause who knows how much extra damage.

As I've said before about smaller autonomous vehicles, they'll be programmed to always choose the option which is most defensible in court. I will be amazed if we ever see an autonomous vehicle on public roads which does something more complicated than "come to a complete halt as fast as possible while steering for the clearest area of asphalt."

There are situations where the safest course of action is to accelerate, or maintain speed, while maneuvering. I'm sure we'll get there if the trend of automating individual driving tactics continues.

"And then, your honour, the truck accelerated and struck a busload of orphans!"

"Case closed."

(I agree but humans are dumb and companies have to cover their asses.)

Robot chooses A) and also immediately uploads footage to the DA Idiot that caused roadwreck is sued for that (I'm pretty sure there are specific laws against causing massive damage. In my country at least, there are.)

Soon, people learn that they are responsible for acting, well, responsibly.

or, apply full brakes, apply full audible warning and swerve at the last possible moment, while calling the emergency services to put them on alert.

But how common is your scenario ?

I consider this a more critical failure :

The Federal Motor Carrier Safety Administration suspects that as many as 28% of commercial truck drivers have sleep apnea.

The Sleep Medicine Division surmised that nearly a 250,000, persons fall asleep while driving each day of the year.

Nearly half of semi-truck drivers have admitted to actually ‘drifting off’ while driving a long-haul route.


> But how common is your scenario ?

Anecdotally, the 3 truckers i've known have complained about it. I now drive very carefully around trucks.

Swerving at the last minute is a non-option because it means turning sharply, which jackknifes the truck. Turning while braking is worse because turning while either braking or accelerating compromises steering. You can try to brake until the last moment and then take off the brakes and try to turn, but again possible jackknife, the wheels could have locked up to the brakes, and now that you're traveling slower you will actually turn slower, so it's harder to cut the wheel and go around. All of this within a few seconds.

You have to either try to slow down and blow the horn, or get out of the way and blow the horn. Some people are lucky and either swerve away from the truck at the last moment, or their car only gets marginally crunched.

One hacker friend of mine's car actually got thrown nearly into the opposite lane of traffic on the highway with the same scenario. Recovery took a while. I think he's still got pain and limited mobility.

Of course, being automated allows for much more advanced braking schemes.

I don't doubt the cut-ins but I mean, how common is the disaster part?

Not as often as truckers having incidents through tiredness etc. I'd warrant.

> You can try to brake ....

We're talking about "The computer ...."

I haven't seen a jackknifed lorry in a very long time, I thought that had been solved with anti-lock brakes.

For this reason alone I would trust autonomic trucks much more than human driven counterpart. I must admit, I've drifted away few times as well while driving passenger car over long, boring highways..

Given that an automated truck will follow a certain safety framework, it's interesting to contemplate if there might be automated truck pirates.

Jam mobile connectivity, run the truck off the road by herding it with other vehicles, then break in and run-away with the goods or the entire trailer for that matter.

You can also do that with a human in the truck. I'm not convinced that it will happen more often with automated trucks.

Most humans will keep driving, even without mobile connectivity. Also, destroying electronic witnesses isn't (yet) murder.

This is one of many reasons we will still need humans present, if only in a guard capacity.

Meanwhile, cargo ship companies are looking toward automation and remote control to get humans off the water. Pirates aren't scared of humans being on board, and in some cases they prefer it, because a human crew is worth more ransom.

Your house sits empty all day and nobody robs it. Video cameras work pretty well for this if it becomes a problem. Great to live in a country with rule of law.

Trucks routinely carry cargo worth many thousand times the value of the average household's possessions.

If a cargo was valuable enough, a driver's life would only be a collateral damage. Furthermore, all (non military or security) personnel training programs I've encountered specifically highlights, that in a case of armed robbery, do not resist, obey and do not escalate conflict in any other manner while trying to call help unnoticed.

Granted, but a human guard nevertheless presents a problem for robbers.

Do you take the guard hostage? Do you kill them? Will they themselves be armed? Will they put up a fight, training or no training? If you try aggressively to bring their vehicle to a halt, will they concede immediately (as an autonomous vehicle likely would), or will they try ramming your vehicle?

A human guard is an unknown quantity, and any crimes you commit against their person are in a different league to the theft of the cargo.

this can be deferred to local security companies with nation-wide contracts.

I think the worry also is that at some point, hundred years into the future someone somewhere might write the code where the truck will go "I'm carrying 1 million dollars worth of cargo which would be destroyed if I drive into the ditch, but the cost of litigation for smashing into this small car that clearly broke the law and will be at fault in this situation is smaller - continue program;".

Don't forget C) attempt B and jacknife your truck across several lanes killing the occupants of the car that runs into the side of your trailer...

'Truckers have to deal with a sort of standard "road safety dilemma" all the time...'

I think you mean that truckers FEEL like they deal with that dilemma all the time. If they ACTUALLY dealt with an A or B dilemma, they would routinely end up in situation A or B: running over a passenger car or flipping the truck. It happens SOMETIMES, but infrequently.

In reality the situations where other drivers do something stupid are probably much better handled by the self-driving truck, which doesn't panic, doesn't feel a fight-or-flight response, doesn't overstate the gravity of the situation... doesn't do anything except assess what to do next without emotion.

I am convinced. Ban trucks.

Mechanical aspects, may be easy. Modeling human behavior and reacting to it, not so much.

I suspect it will still be years before much of trucking is automated. Why? It's not just the take crap from point A to point B, it's also being responsible for making sure what was ordered is actually loaded. It will be years before, say, the laborers at Earth Bound Farms south of Salinas (supplies large quantities of "organic" produce all up and down the west coast) automates their work staff or actually has their schedules down tight.

Trucking and hauling is full of inefficiencies, and those that suffer the most are the drivers. A pickup/dropoff that takes too long because the people at the origin/destination are lazy/slow/don't care/behind has a ripple effect.

And, one could argue "but automated trucks, what schedule?", there will still need to be a human present, regardless. I suspect said human will be restrained by rules similar to what drivers experience now.

> the laborers at Earth Bound Farms south of Salinas (supplies large quantities of "organic" produce all up and down the west coast)

Why on earth did you pick them out so incredibly specifically and put scare quotes around organic? Doesn't seem relevant.

But we've already got automated warehousing systems that work pretty well -- I'd imagine these wouldn't be too hard to extend to load and unload trucks. In fact, I wouldn't be surprised to find out that they already can.

Specifically, the automation of long haul trucking, comparable to what large tanker ships do, may be more feasible. The last mile, scheduling, and the rest, is a horrid mess. A family member is in the business and I live in an area where ag work is prevalent and know people who have worked in such.

Agreed here. I don't think anyone reasonable is suggesting the local delivery trucks are going to get automated any time soon.

But the long-haul interstate runs? I see this happening faster than any automated passenger cars, and once it starts it's going to rapidly take over that industry. Probably starts with convoys with a couple human tenders until the tech is proven.

Where is the money going to come from for produce suppliers, some on thin profits, to automate where hiring a bunch of seasonal laborers is less overhead, maintenance, etc.

It's going to come from automation costs dropping and seasonal (and other) labor costs rising in real terms, so that the cost effective approach is automation. Same as everywhere else.

if an automated solution ends up being cheaper than manual laborers, then some lending institution will be happy to lend the money for the transition. any players that avoid the transition will likely get obviated by the competitors that capture and pass on the cost savings.

It is coming from another Uber company. Or maybe a new brunch of Uber.

It won't be a perfect fit for all types of trucking (say, delivery drivers), but many loads can certainly be transported place to place without a person needing to babysit it. Example: I imagine Amazon would love to be able to move loads around distro centers with automated trucking.

> Things involving humans (vehicles cutting you off, pedestrians, etc.) are part of the general class of problems that I'm assuming all automated vehicles are working on and are much harder.

The more mass you have, the longer your braking distance and the more sluggish the reactions. While the problem is the same, and the mechanical differences easily modelled, light vehicles get away with lower accuracy and higher latencies, because they don't need to plan ahead so far.

But it's still simple math. More weight, be more aggressive with brakes.

If human truck drivers can do it, then a computer truck with a 360 model of the environment around it can do it more efficiently right.

I don't understand the problem here really. Computers are way better at logical computation than humans and that's all we require here.

Yes, but there are more problems to consider here. With some types of cargo, applying maximum brakes can actually damage it(think livestock for example). You also don't want the truck to be going into panic mode and applying maximum brakes for a plastic bag or a small animal. I'm afraid that our current technology can at best tell that there is something there, but it won't be able to tell what - humans at the moment are much better at visual recognition. It's just a concern at the moment, but I think it's an important one because it can lead to some serious accidents when the technology launches.

have you not been following the developments of self driving cars? Like the story of how google cars had problems gauging the intent of bike riders that were themselves stopped to let the car pass ... it looked like they were about to take off, so the car was too conservative and didn't move until the bike changed what they were doing. And then, once that condition was experienced and identified, it was fixed for all google cars in the entire fleet.

I'm pretty sure the car will be able to identify and plastic bag ... not to mention a simple cargo manifest that would apply parameters about how the cargo is to be dealt with.

My point is that all of these are "simple" problems. Oh the truck will identify a plastic bag. Oh the truck will have a cargo manifest and drive accordingly.

But these are all new technologies. Our visual recognition libraries are nowhere near ideal. I said this elsewhere but I will say it here too - our best of the best software can't tell a zebra and a sofa in a zebra print apart. I'm pretty sure that all that LIDAR is seeing is a spherical object on the road, it has no idea what it is.

The same with the cargo manifest - it sounds simple, but no one has a system like this. It would need to be made and implemented, and it takes time and effort - I'm not saying that it's impossible, far from it, but it's one more obstacle to overcome to make this possible.

> I said this elsewhere but I will say it here too - our best of the best software can't tell a zebra and a sofa in a zebra print apart.

Really? I find this surprising. Do you have any links or sources? I'd like to learn more.

There was a few articles about this exact issue on HN previously, a cursory search returns at least one of them:


Fascinating, thanks for sharing!

Are the lower masses not offset by higher speeds though?

The article mentions 45mph as the most efficient speed for a truck. I can imagine the regulators demanding a pretty low speed for automated trucks.

I wonder what the accident rate of professional truck drivers is compared to other passenger class drivers. Certainly the outcome of an accident is potentially a much higher level of mayhem.

Not exactly what you're asking for, but related. In accidents where both a large commercial truck, and a car were involved:

"the car was the encroaching vehicle in 89% of headon crashes, 88% of the opposite-direction sideswipes, 80% of the rear-end crashes, and 72% of the same-direction sideswipes."


Now you need to take into account how many cars vs how many trucks were on the road in the period that yielded those numbers. Only then you can obtain the propensity of car drivers vs truck drivers to get into accidents.

Not sure I understand...these are only accidents in which both a car and truck were involved.

The definitive thing you would want, I assume, is "at-fault accidents per xxx miles driven" for all accidents, segmented by vehicle type.

Yeah, that one makes sense. Because if 1 truck has 30 cars around it, chances are that a car will screw up before a truck.

Sure, but unless the car running into the truck _prevents_ the truck from being at-fault in a different accident further down the road, that doesn't really matter.

> And then when you get to your destination, manoeuvring a 50' trailer is no easy task, even for a trained driver like my dad.

I wonder if there will be a more specialised role for human drivers here akin to pilots in ports. Companies at the scale able to invest in the tech might have the trucks arrive at a queue near regional hubs and have wet-ware do the 10 minutes of fiddly driving after clearing off the Smart car debris.

The article says so. The plan is to have human drivers in each city drive the "last mile" from the interstate freeway exit to the destination. The current autonomous technology is only effective on freeways, so having human drivers drop the truck off at the interstate, and pick it up again solves the temporary problem of not having 100% effective automation of city driving/trailer parking.

If I were younger and not married, I could totally see taking a job as a trucker and guiding it on and off the highway, and doing coding or whatever in the meantime. It'd be a chance to go different places and have some adventure, while continuing to do another job that I love.

It's not unlike many college kids taking a job as a security guard and "watching monitors" while doing their homework, etc. (Another thing that I considered, but never actually did.)

Of course, there's a lot more training for truck driving, but I don't see that as a major barrier, given the benefits of extra pay and travel.

You don't get the point. You won't be traveling inside that truck. You will only drive trucks in and out of your city.

I see the evolution of automated vehicles as slowly taking more and more mental load off humans. e.g

1) Automated gears and obstacle detection - we already have this. 2) Have the truck stay in one lane and maintain distance - We're close. This means humans stay in truck and solve the hard problems when encountered. (Kind of like pilots) 3) Automated port to port driving without humans. 4) Destination to destination driving without humans.

I bet the human driver doesn't even need to be physically in the truck.

We already fly military aircraft remotely, I don't see why a truck that's packed with cameras and sensors can't be driven safely by a human in an office somewhere.

I wouldn't be surprised. This is where a huge amount of real-world automation ends up; it ends up being much more practical to only automate the 90% of the job that can be done by a simple automaton, and mechanical-Turk the rest. Then you save 90% of the labour costs AND 90% of the R&D costs for the automation.

If I had to bet, that's the way I see it playing out.

In the states, many distribution centers are located next to major transportation arteries, whether rail or interstate highway. In addition, many of the places they ship to are also located on major arteries. It wouldn't be a huge stretch to have automated trucks you load on one end, they drive to the other (using 98% highway driving), then park in a waiting area until a human finishes the job by parking them.

The interesting edge cases here would be accidents, construction, and detours. Whatever you deploy should be able to handle those without human intervention. Otherwise we could end up with every accident site also having 100 automated trucks at a dead stop waiting on drivers to show up. That'd be ugly.

Are you sure they're more complicated for computers? Because some stuff that strains humans, like 360° attention and complicated sequential activities, computers can do with ease. Also computers are not limited to having senses inside the cab. Load shifting? BigDog is an example of computers figuring how to balance a shifting load. Google cars handle being cut off (most of the time).

Google cars are going no more than 25 mph and are lightweight. Different thing all together.

Here is the big thing - the driverless truck has to operate in a uncontrolled environment that can result in human death. Who is the responsible party? The trucking company? The software developer?

Additionally we are talking situations where a computer can't have any reaction interruptions for hours and hours. No reboots. No garbage collection pauses. No panics.

Can this be done? Maybe. But at what cost? It might be that the best role for the computer is helping a truck driver be safer -- not to replace the truck driver.

Additionally we are talking situations where a computer can't have any reaction interruptions for hours and hours. No reboots. No garbage collection pauses. No panics.

You mean, just like existing ones? It's not like we're still driving purely mechanical cars. Current production vehicles already have plenty of software managing stuff.

I'm walking down a busy road with traffic passing me at ~30mph at the moment. Anecdotally, the last 4 cars have had one person in the car who has either been texting on a mobile phone, applying makeup, eating breakfast, and playing with the radio. Do you think the reaction time of the guy who's texting his boss saying he's late would be faster or slower than the driverless car that was in a long garbage collection, if I were to step in front of him?

You do have a point on liability, and I don't have an answer better than a centralised pool of money that everyone pays in or out of.

Yeap. And if one of them hits you, they are liable. How many software developers want to be held criminally responsible for deaths?

Some times we get so fascinated by technology, that we forget that the technology has to fit within human society.

Remember the whole concept of product/market fit - if the market rejects the product: it doesn't matter how "cool" the technology is.

>How many software developers want to be held criminally responsible for deaths?

Nope. Short of gross negligence in design, that seldom happens in any engineering field.

How many people were criminally charged in the Challenger disaster? None.

7 people died in the challenger disaster. 30,000 people die each year in car related fatalities. Completely different order of magnitude.

Additionally, going up on a rocket is a fundamentally risking proposition done by adults.

Having a driverless vehicle that glitches, applies power instead of brakes, and kills a bunch of pretty young girls.... well that makes the 5 oclock news and 11 oclock for many, many days.

People get charged with crimes.

How many criminal charges were filed against individual engineers for any those 30,000 fatalities last year? How many were convicted? None and none.

You're illustrating my point for me quite nicely. Frankly, I can't think of a criminal conviction, even at the corporate level, for design flaws in the auto industry without going all the way back to the Ford Pinto.

I wouldn't be surprised if they start out with "last mile" drivers to deal with off-highway driving.

This was my thought as well. That a new 'tugboat captain' or 'harbor pilot' type role emerges for last mile driving.

But here’s the question: Do these ‘last mile’ drivers sit in the cab? Or operate the rigs from thousands of miles away, like drone pilots, with displays showing 360 degrees around the rig and access to all of its sensors?

Last mile driving may not look anything like truck driving.

Or, alternatively, do they wait at truck stops on the edge of town and meet the trucks as they come in (much like harbor pilots meet container ships at port entrances)?

If you were able to dispense with the cab and any human interfaces/amenities, that would cut out a big chunk of weight and drag. It would look radically different but just think of all of the components on a truck that are only there for the driver. No need for a cab, AC system, heater, pedals, steering wheel and linkages, airbags, seats, pretty much all of the glass and windows, doors, etc. The added sensors, telemetry, and computational requirements necessary for an autonomous vehicle suddenly don't look so pricy compared to the savings of removing a huge chunk of the vehicle, nonstop operation or maintenance, no salary for the driver, lower insurance costs, lower fuel costs, and other assorted savings. Really it's looking like more and more of an inevitability in my opinion. Once the control problem is solved, I doubt it'll be more economical to continue putting a trucker behind the wheel.

Of course if you need to provide all of that to allow for someone to drive it manually around town a lot of those benefits are gone. Who knows, there will probably be some transitional period where all automated trucks are still equipped for manual control but I doubt it'll stay that way for long.

Even if we remove all the facilities for a driver, i likely you would still need a human guard.

Although thinking about it, the guard could travel in a separate vehicle, and could be responsible for overseeing an entire convoy of trucks at one time.

I don't know what became of it, but at one point there was a DARPA initiative to do just that for military convoys. The reasoning was that automated trucks would allow soldiers to treat the convoy as a "dumb herd" that would follow and be overseen by a squad of heavily armed and armored "sheepdogs". This would improve protection for soldiers, while simultaneously improving the efficiency of battlefield logistics.

The automated part of the journey could be done by a driverless / cab-less / HID-less prime mover which drops the trailer(s) / cargo containers off at an interchanged where they are picked up by human driven vehicles for the last mile.

I don’t think so. Once you’ve already automates a truck, you have a lot more information to convey than what’s out the windshield, and you are liberated from needing to turn a steering wheel or push foot pedals.

There’s no reason to build a cab into the truck that is only used for the last mile. Telepresence will do just fine.

If that's the case, then why aren't cargo ships already automated to this extent? It seems like automating a container vessel would be even easier - there are far fewer things to run into on the water (even in a busy port) than there are on a busy highway.

I would bet because the wage paid is insignificant compared to maintenance of the vessel and that the maritime laws would be further complicating such a scenario.

Not so much maintenance as capital costs, I think. Giant cargo vessels can't be replaced as often as trucks either. But drone ships are certainly coming, it's a very trendy topic in the trade publications. I'd give it 10 years.

Pirates? You need a crew on cargo ships regardless of whether they are automated or not.

Crews generally aren't armed to begin with and with no humans to capture and ransom or threaten to change the ships course the amount of booty they could make off with is really limited. Most pirate attacks target hostages anyways. For engineering you could have small crews fly out to any one ship that's experienced engine trouble.

Definitely can't remotely operate a 40 ton truck. Wireless network latency is a bit high even under ideal conditions, and latency spikes could easily cause disastrous accidents.

You can continue to rely on the automated systems for avoiding collisions. The intelligence is more with dealing with the decisions that the AI can't make.

If you can pilot flying killing machines half the world away, you can definitely control a slow moving truck remotely.

There's not much to run into in the air...

I don’t think you’re going to be “driving” such trucks. I think you’ll be making decisions. For example, when you reach a yard and need to back it into a loading bay, you may draw a path on a map of the yard, taking into account obstacles that are there at that moment. The truck will handle the turning, backing up, and so on.

Likewise, you may be monitoring city traffic and choosing alternate routes. But you won’t be driving with one foot on the brake pedal and your hands on the steering wheel.

You could make massive gains just by not having to stop and rest drivers. I suspect the drivers will be on-board but able to relax or sleep until getting close to the destination. That lets all trucks drive 24hr straight through to their destinations.

>Or operate the rigs from thousands of miles away, like drone pilots, with displays showing 360 degrees around the rig and access to all of its sensors?

Any driverless technology that requires a constant wireless internet connection is a non-starter.

As I said elsewhere, I do not think a constant wireless connection will be an issue, because I don’t think you’ll be operating the truck remotely the way a hobbyist operates an RC car. The truck will brake when a car, pedestrian, or bicycle cuts it off, and do it better than you could.

You are there to route it around traffic jams, talk to the foreman on site, and choose the loading bay.

Even if you had a reliable high-bandwidth/low-latency connection, the truck will be better than you are at handling things that requires instant response. You are there to make decisions.

That's true, it requires a higher level of skill from the AI though.

The video stream bandwidth and requirement for near-zero latency will make remote control very challenging, if not impossible. I'd say the last-mile driver will be in the truck.

But the last mile is where driverless trucks are most useful. For long haul, you can just use trains.

Actually you can't, for a lot of goods. Reason is that the railroad won't give you the same guarantees and trackability a truck will.

Example: you have refrigerated items (let's say a shipment of fresh fruit) that has to be maintained at a constant temperature and you want it delivered to your customer on Thursday when you ship Sunday night.

Truck: not a problem, and you can easily get updates all the time from dispatch (or if there is a GPS unit on the truck, immediately).

Railroad: maybe they will get it there, maybe they won't.

If the railroad puts a refrigerator unit on your boxcar and it breaks down, too bad - they won't reimburse you. A trucking company will.

> Railroad: maybe they will get it there, maybe they won't. > If the railroad puts a refrigerator unit on your boxcar and it breaks down, too bad - they won't reimburse you. A trucking company will.

Sounds like both of those are more to do with shit rail companies than inherent downsides of rail (or upsides of trucking)?

Just in time delivery by rail works in Europe, perhaps because the track is maintained to a high standard and can be relied upon, and the timetable adhered to to avoid delaying passenger trains.

> If the railroad puts a refrigerator unit on your boxcar and it breaks down, too bad - they won't reimburse you. A trucking company will.


> Insurance?

If the thing you insure against actually happens, premiums go up to the point of it being uneconomic.

Sorry to burst your bubble here -- but the railroads were doing exactly that since the days of refrigeration via blocks of ice. ( http://californiabountiful.com/features/article.aspx?arID=56... )

> Now retired after more than 30 years in the produce business, Wolf noted that farmers began growing iceberg lettuce in the region just prior to the 1930s. The demand for the hearty lettuce variety soon took off through the use of refrigerated rail cars.

> But the last mile is where driverless trucks are most useful. For long haul, you can just use trains.

And yet the highways are full of semi trailers...

IIRC because more of the highway cost is subsidized by the taxpayer, not necessarily because the cost is actually lower...

In the US, especially outside of the northeast, most interstate highways are not tolled, so users of the road don't pay any more for its maintenance than non-users do. But someone is still paying for it. Taxpayers who don't use the roads are effectively subsidizing those who do.

By contrast, railways usually own their rails and are obliged to maintain them themselves, and thus have to fold those costs into the price.

This is really unfair.

> For long haul, you can just use trains.

RoLa (Rollende Landstraße) as we call this in Germany, or just transporting the flatbed carriages, requires a SHITLOAD of manual work, unlike containers where you don't even need people on the locos in the yard anymore.

Not to mention that long-haul-via-rail is only viable in the US... in Europe, it virtually doesn't exist anymore. It just isn't profitable due to various external conditions:

1) train track pricing is prohibitively expensive

2) freight trains don't get priority over passenger trains - quite the contrary. If you have any time-sensitive stuff, better transport it via road instead.

3) rail tracks are massively overbooked. One single delay and the entire system grinds to a screeching halt.

4) there is no such thing as cross-Europe locomotives/conductors. There might be multi-system-capable locos, but there is not a single loco on the market able to serve the entire fragmented European railroad technology.

There may be "just" 15/25 kV AC + 1k5/3kV DC, but every country needs their own roof collectors because they're NOT compatible (some countries don't even have roof collectors, but 3rd rails or sideway rails for the low-voltage DC stuff, and the DC stuff usually doesn't pack a lot of power so you can't really use them for any heavy load), but the real tech problem is the load of bollocks called signalling/security system, every country has their own, including Germany alone with FOUR popular systems (classic form signals, light-based signals, LZB/PZB and ETCS, in addition to various customizations in Eastern Germany and the S-Bahn tracks).

Not to mention the conductors which have to know all the signalling systems AND all the languages (compared to aviation where English is the standard worldwide), and they need to be expensively re-certified...

Yeah, ETCS, I hear you, but ETCS is a joke that would require literally hundreds of billions to retrofit across the major transit routes in Europe, but that won't happen - we Europeans haven't even managed to get our freight carriages to silent brakes yet!

5) There are still restrictions like maximum train lengths and the plain fact that European freight carriages only carry a pneumatic line for the brakes, but no data lines for controlling middle/back locos - so there's no technical way of extending train lengths because there's simply no loco combo capable of pulling >700m length heavy freight carriages with 3 locos on the front and no way to control a pushing loco in the back of the train. Simple physics, the good old screw coupling just can't support more load. There are alternatives (C-AKv coupling, it's high load AND compatible with the old coupling) but once again a retrofit would cost billions (see #4)...

Trains might be more efficient from an environmental and human-resources POV, but they're fucked up on any economic scale - and politicians all over Europe don't give a single fuck, instead they are all letting their rail systems rot to pieces and squeeze every tiny penny from the rail companies to fix up their even more rotten state budgets!

> RoLa (Rollende Landstraße) as we call this in Germany, or just transporting the flatbed carriages, requires a SHITLOAD of manual work, unlike containers where you don't even need people on the locos in the yard anymore.

Isn't using containers much more efficient anyway, instead of moving the whole truck?

> Isn't using containers much more efficient anyway, instead of moving the whole truck?

The core problem is that freight companies in Europe standardized on the Euro Pool Palette (EPAL) but ISO containers worldwide use the non-metric US measurement system - thus leading to waste of space and potential safety issues (sliding palettes due to slack space on the floor).

Trucks, however, are fitted to the metric system and can safely transport EPALs, so you've got next to no alternative.

ETCS is coming - it may take twenty years to be rolled out everywhere but it will happen. In the meantime they're focusing on the strategic routes first. These things take time, but it is happening.

> 2) freight trains don't get priority over passenger trains

Interestingly in the northwestern u.s. (and I suspect throughout the western u.s.) the main rail corridors prioritize freight.

Discovered this while, as a regular Seattle - Portland traveler, I found myself stationary while lumber rolled by. Amtrak's leases are subordinate to the freight lines.

> 2) freight trains don't get priority over passenger trains

Freight trains get priority here and it still doesn't make sense to ship by train. Rail is pretty limited by the huge size of our country, and they need to be able to ship to tiny random stores all over the place. It makes more sense to send one truck that can stop anywhere there's a road than send it out via train and then transfer to a local truck, which might be sitting unused the rest of the time.

> Rail is pretty limited by the huge size of our country

Quite the contrary, the US is better suited for rail than the rest of the world, at least for cross-country moving of goods (one single area of law, no customs crap to deal with, the only thing I'm not sure is signalling and interoperability of locomotives on rail systems in different states).

Trains are perfect for inter-city travel, the first/last mile should be handled with trucks unless the facility is so large that it can source or sink a huge number of trains (e.g. automobile, steel factories, coal mines, fossil power plants).

How long do you imagine it takes to unload a single car or two cars in a given city and move on to the next city? Once that "last mile" is handled, what else is that truck going to do until the next train shipment arrives? It'll sit doing nothing. If it moved on to another city, it could have just been loaded somewhere else and didn't need to be loaded at last mile.

Trains have several single points of failure for a larger load. When there is a problem, all of the load is affected, it takes longer to repair, and problems with both the rails and the trains happen regularly. And trains have more complex maneuvering than trucks; switching lines is a non-trivial time-consuming matter and operators have a tough job just on the existing lines.

The fact remains we don't have enough rail to service most of the country, and nobody is paying for new rail, not to mention the issue of fitting more track on the existing land and bridges used for rail. Demand has been increasing steadily on rail, and only so much traffic can possibly run on the rails - which are shared with passenger trains, btw. To switch to trains you'd be adding 230% more traffic to the rails.

> Once that "last mile" is handled, what else is that truck going to do until the next train shipment arrives? It'll sit doing nothing

If you're a responsible freight truck company, you want to avoid this situation at all costs. There's always cargo that wants to be shipped - the only scarce resources are drivers and tractor units, and some companies even let multiple drivers share the tractors to maximize the usage of their tractors (that really depends on the company though, there are also companies where every driver "owns" his tractor).

> The fact remains we don't have enough rail to service most of the country, and nobody is paying for new rail

Yes, indeed. The beginnings of the stock markets actually were railroad construction companies, but these days the financial markets only care about the next big unicorn and not about creating something with real long term value, which a train track certainly is given that, once the track is laid, the land belongs to the company and the only major investments required are 50-year overhauls and regular maintenance. It's a shame.

Tell that to the companies that ship items long-haul via trucks. It confuses me too, but somehow, I guess the economics work out.

In Australia this isn't the case - the infrastructure doesn't service a lot of the land area.

Trains can't haul produce and other refrigerated product.

The Juice Train can, as does the "Salad Shooter" train. https://en.wikipedia.org/wiki/Juice_Train

It actually talks about that in the article:

We still need to create on- and off-ramps so human drivers can bring trucks to the freeways where highway autopilot can take over.

I think the flaw here is assuming that the systems around the truck will stay the same - including the truck form factor. Without the required creature comforts of the tractor, you free up a lot of energy (heat/cool) and space (bed/TV etc...) in the cab.

I can also see traffic lanes being created to support commerce, special easement areas for trucks at stops that don't have to support showers/food/bathrooms etc... reducing overall cost.

All of this comes at cost to human workers - but it's inevitable.

> All of this comes at cost to human workers - but it's inevitable.

I disagree. Specifically, you are assuming that the human workers will quietly accept their fate.

Not quietly, no, but it's no less inevitable.

History is full of jobs that have been automated away, and few of those historical examples offered a more obvious and large payoff.

I do notice you are avoiding my central point:

With railroads, there is not an economic incentive for automation.

One commentator suggested that driverless trucks would have to be separated from human drivers. Railroads have this thing called "tracks"

Another suggested the wonderful advantage of a driverless truck not getting into accidents. Railroads have an excellent safety record.

Driverless trucks as a way to reduce the need to spend money on humans. Railroads need only 2 humans for 100+ unit train that is equivalent to 300 or so trucks. ( 1 railcar = 3 truckloads https://www.chrobinson.com/en/us/Logistics/Intermodal-Rail/C... )

Trains are 10x more ( http://grist.org/article/freight-trains-19th-century-technol... ) more energy efficient than trucks.

What is the economic model to use driverless trucks? Sure the trucking industry will push for this. But the railroad industry has so many natural advantages.

What is the economic model to use driverless trucks?

The same as for trucks with drivers (only cheaper) and those already control much of the transportation business.

>What is the economic model to use driverless trucks?

Road goes to more places than train tracks.

You dont really need more than that...

Butlerian Jihad! You shall not have thinking machines.

"I think the flaw here is assuming that the systems around the truck will stay the same"

That's an interesting point. I wonder if self-driving trucks will tend to be smaller, because there's less incentive to amortize the wages of the driver by making his/her truck as large as possible.

The geometry- and mechanically- related tasks seem like they would be ripe for automation. For example, the trailer reversing problem is related to the inverted pendulum problem, which is easily solvable by automated systems.

The judgement for when to stop with an unsafe load and wayfinding at the last 100 yards of a route are indeed tasks that I'd prefer to leave to a human for the time being. Maybe we'll have 'dock pilots' like we do for large cargo ships?

Build a robot that balances this broomstick was a sophomore-level engineering course at my university. xD

We've gotten used to huge commercial airliners flying on autopilot. Granted they have humans on board, but ask any pilot how much flying they do. Even helicopters can be put on autopilot after take-off. That's what the helicopters that run from Macau to Hong Kong many times a day are doing. I can see ironing out the kinks - pedestrians, roadblocks, car stops in front abruptly, etc... I would think you could have a truck lane only like the ones for buses entering the Lincoln Tunnel in NYC, and perhaps a guardrail that could have sections removed to open it up for traffic during less traffic times?

Honestly, I'm not sure that most people understand the extent to which commercial airline flight is automated. For example, most people are shocked when I tell people that pilots are required to let the plane auto-land under very bad visibility conditions.

> "most people are shocked when I tell people that pilots are required to let the plane auto-land under very bad visibility conditions."

As a commercially-rated pilot I would be shocked, too - since few airport runways are certified for Cat III landings, and that's only under restricted wind speeds.

And even after autoland in fog, a pilot still taxis up to the gate.

Jeez, I admit I am, was now, one of those people!...changes flight to train ticket... ;)


They may have humans do all the manoeuvring within the distribution hubs long after they've automated all of the street driving.

Shifting and braking, though, seem to me like the kind of thing that automata should be good at. And I suppose that, once most of the cars on the road are autonomous, the streets will be much more truck-friendly.

Sort of reminds me of a Harbor Pilot[0]. A person who holds unique knowledge of how to operate a large vehicle in a particular piece of confined geography.

0: http://www.insidejobs.com/careers/harbor-pilot


I doubt a "hub driver" will earn a 6-figure salary, but manoeuvring a semi truck two or three trailers is definitely a skilled job that many (most?) people are not capable of learning to do well.

>>Trucks are way more complicated to drive.

An Australian friend of mine told me that the trucks that carry load from Australian mining fields are almost fully automated, and have been for a while now. So perhaps it's not that complicated?

They may have been thinking of trucks within the mine site, which is still quite new? (http://mobile.abc.net.au/news/2015-10-18/rio-tinto-opens-wor...). I'm pretty sure we don't have any automated trucks on Australian roads.

This is mostly internal to mine sites, and the areas serviced have restricted or no public traffic.

The article actually mentions that initially we will need human drivers to get the trucks on and off the freeway:

"We still need to create on- and off-ramps so human drivers can bring trucks to the freeways where highway autopilot can take over."

I was referring more to the inner-city manoeuvring, and within loading/unloading zones. Getting a trailer lined up with the loading dock isn't an easy task at all. It can be automated, but it's the variability between load length, weight, and the loading dock's surroundings that make it difficult.

A comment below talked about harbour pilots. That seems like a great idea.

But my point, and the one made by the article as I understand it, is that you wouldn't even have to do any of that initially. The vast majority of the driver's time is spent just cruising on the highway. That's the part you want to work on automating first, both because it provides the greatest cost savings, and because it's the easiest to solve technically.

Lexus automated the most hated driving maneuver for all car drivers - parallel parking, and Google's had decent success w/their self driving Prii fleet... it's hard, safety would be the #1 concern, but w/enough human brainpower and some decent hard realtime OS-es, it should be more than doable.

> I think people will be less comfortable with the idea that the truck next to them has no human in it, and could experience some kind of glitch with catastrophic consequences.

Unless the truck somehow acknowledges adjacent vehicles in a comforting way. I don't know what that would look like, but it would probably be coupled with advertising and data collection (no law against filming all the cars around you and what they are doing).

At least long-distance trucks can have routes optimized for automation (be it via embedding sensors to aid in bad weather or optimizing a lane for trucks, etc.) for the long segments and then at the city limits (or thereabouts) set-up truck "depots" to have human drivers board the truck and give them override to manoeuver through more difficult city or rural last-segment roads and streets.

>>And then when you get to your destination, manoeuvring a 50' trailer is no easy task, even for a trained driver like my dad.

My first airplane journey was fascinating. As some one who doesn't know how to drive a car, and is struggling trying to learn it, I was spell bound when I watched the pilot taxi the huge airplane to the terminal.

Well airplanes aren't even regular shaped and size of the human taxi'ing the plane to the plane is very small. Yet the pilot could precision park to the terminal.

The fact is no matter how hard it seems, if you've been doing it day in and out, it looks like just another job at the end.

I think it could help that loading docks could be highly integrated with the truck systems. Backing in to the place where you unload is complicated, but the owners would be willing to invest in putting in sensors, special paint, whatever else is necessary to make that as automated and error free as possible.

If everything was smooth and regular, trucks wouldn't have a problem. However, the roads, and the warehouse yards, are not.

A fully loaded truck traveling 55mph (on some roads they are allowed 65 or even 75) can weigh 80,000 lbs or more (in Canada under certain circumstances, 139,500 lbs).

That is a tremendous amount of force that must be handled properly 100% of the time, through rain, hail, sleet, snow - and with smaller cars weaving in and out of traffic all around them.

It's already not handled 100% of the time. I'll settle for 75% less freeway blocking by crashed big rigs.

I am hopeful that robots can fix it, but most likely it will take a little bit longer... robot trucks will save on fuel, brakes, and tires, as they can be programmed to minimize sudden acceleration and taking a turn too quickly (scrubs rubbers off the tires).

Having driven in winter weather I can honestly say it will be a while . . before they are ready to take over. As for tire wear , there are small firms that look for very experienced driver's to cheat the scales . . this is how its done if they are overweight , think ( steel coils ) they time it so they hit the scales at night . . the driver goes into the scale and positions the rear trailer wheels on one side . . up on the curb registering a lower weight on the scale.

The demonstration in Europe shows that driverless trucking is right around the corner. The primary remaining barriers are regulatory.

What about the roads themselves? I'm curious how an autonomous truck will handle a snow-covered highway, fog, potholes, and tight corners caused by narrow roads, bad parking, snow banks, etc. This is a way of life where I live (Boston area).

The article about the Russian driverless van (cited by @ommunist) talks with the experts, who are by no means 100% convinced this is "right around the corner": (1)

Industry players are aware that just one serious accident on the roads involving a driverless car could set back their development by decades.

“Everyone has come to understand that the technology is 90 percent ready, but even when it’s 99.99 percent ready, it still won’t be launched,” says Gol, whose product is currently being tested in various scenarios and weather conditions.

“Until we can teach [artificial] intelligence to learn the nuances of things that can occur unexpectedly on the roads … what will happen if transparent glass is being carried across the road, what if there’s a paper bag that the system perceives as a rock,” the system isn’t ready, he said.

“We need a mathematical revolution to overcome that 0.01 percent,” says Gol. “We all understand perfectly that to have just one accident involving autonomous transport will have a huge fallout in society – the technology could be closed down completely for another 50 years. It’s a huge responsibility,” he added.

1. https://ninja.oximity.com/article/Russia-ready-for-driverles...

"What about the roads themselves? I'm curious how an autonomous truck will handle a snow-covered highway, fog, potholes, and tight corners caused by narrow roads, bad parking, snow banks, etc."

It won't; at least, it won't need to handle narrow roads, bad parking, snow banks, etc.

The autonomous truck need not drive the whole route that a human driver does now. Drive down the Ohio Turnpike and you will see numerous parking lots with UPS trailers parked in them, right alongside the highway. All you need is an autonomous truck to drive the trailer down the turnpike. Humans can drop it off at the turnpike and pick it up. It could deal with typical weather events, or just look at the weather forecast and stay parked if the weather looks like it will be bad.

The highways are cleared shoulder to shoulder a day or two after most snows so the autonomous trucks wouldn't lose too much yearly time due to bad weather. You could keep humans around to drive the more urgent shipments.

> The autonomous truck need not drive the whole route that a human driver does now.

Exactly. They'll be more like trains than trucks as we know them; they'll simply have more tracks available.

Makes sense. Do the hundreds of straight highway miles driverless, then transfer to human driver for tricky last bits. Dumb question - why not use trains?

Would it still be an advanced net if you had humans in the trucks as a 'backup', with the autonomous driving doing 90% of the 'boring' driving down the highways? Humans would still drive the last mile (ha!).

Would that be significant safer and more efficient to make it worthwhile?

If you limit autonomous driving to the main freight routes on the highway, I can see self driving vehicles happening a lot sooner than I previously expected. Like 20 years sooner.

I think that will be the near future; we've already done the exact same thing with airline pilots, who spend most of a flight on autopilot (hence the expression). This allows them to do long shifts with much lower mental pressure.

I'm even imagining 24 hour drivers, who let the truck drive autonomously at night on quiet roads at low speeds while they sleep and live in the truck. A bit like lonely spaceship pilots from science fiction.

i think this is a likely first application. you can boost schedules and efficiency massively by simply allowing a truck to drive itself to the next waypoint overnight (maybe not even the whole night -- just to the next safe stopping area) when the driver would normally be sleeping. if there's an emergency, simply stop and pull over and alert the driver.

this is probably also much easier to sell to existing industry and government. you're not eliminating drivers at all, but instead making their jobs easier and profits higher. having zero autonomous trucks on the road during the day is also a PR positive. no road tripping families put at risk, etc.

I can't wait for the day that putting 0 human drivers on the road is a PR positive, because no road tripping families are put at risk.

not going to happen in the US. you may not like it, but that has nothing to do with reality.

Reasoning pls

They started truck automation in mining and container terminals in ports. Areas with no other traffic and the possibility of active guidance (beacons along/embedded in the route and so on.)

Highways are the next obvious use. I could see automated trucks driving down the transcanadian highway for example or similar long interstate stretch.

I gotta say though, if we're going that way, why not go intermodal, add a RR right of way through the interstate with a double-track, operate automated freight trains and automated transshipment terminals. Just load the trailer onto the train and go. I know rail isn't sexy but it'd make sense in terms of carbon footprint (trains could be multimode diesel-electric/electric depending on corridor.) And most of that rail net could be fully automated.

Rail suffers a lot from inflexibility and the massive redundancy required to survive the occasional track maintenance. Trailer-on-railroad does not really improve on that. A true logistics moonshot would have to go all the way to "roadgoing, powered railcars" that could platoon on that double-track and maybe opt into some form of power delivery (overhead line?) where available, switching on the fly between rubber/air and steel wheelsets as required.

The beauty of this expensive approach is that it would not only obviate waiting for a train going in the right direction and the requirement for mode switching terminals (those are just the fair weather sailing benefits), but also eliminate the inherent fragility of highly interdependent railroad schedules. Where conventional rail quickly breaks down completely, road can gracefully degrade towards a massive traffic jam slowly grinding forward on some dirt road fallback route.

Roads are really expensive. Adding rail lines is cheaper to build and operate so a dedicated automated truck lane seems like a poor idea. If anything building 100+ MPH rail for long trips like this seems like a great investment until you adjust for actual traffic levels.

They don't need to build new roads, they just need to add cones to mark off a lane on an existing highway.

> If you limit autonomous driving to the main freight routes on the highway, I can see self driving vehicles happening a lot sooner than I previously expected. Like 20 years sooner.

It is 100% certain without any doubt whatsoever that this is how it will start. It will start with dedicated interstate roads (or lanes) just for driverless trucks, and then cars.

After some time all freeways will be 100% driverless. I don't anticipate mixed use (driverless + drivers) ever happening.

Local roads (i.e. uncontrolled access) is many many decades away.

> I don't anticipate mixed use (driverless + drivers) ever happening.

but it already has?!

I mean there are already self driving cars being tested on public roads. Technically not driverless, but almost.

Those cars are not able to handle general purpose driving.

Say you are on a one-way road, and at the bottom of it an ambulance is blocking the road - so you need to back all the way out of the road (or u-turn).

No driverless car today can handle such a situation. There are many more real-world examples. The cars today are just demos, not usable products.

Can you elaborate on why you don't see mixed use ever happening? Do you think its regulatory, liability or technological... or maybe some combination of the three?

Because it will never be needed.

The steps go like this:

Driverless car in a dedicated lane. That will be first since it's much easier to build than local driving or mixed long distance. It also covers the most important need (i.e. much less economic reason to make one for local use).

Lots more people see how great that is, so more cars (and trucks) will be rapidly added. That lane will reach capacity, so more lanes will be added.

Within 5 to 10 years driverless long distance will be the norm, and long distance driver cars will be rare. (All the cars will be have mixed modes, with the driver taking over for local streets.)

Every new car will advertise that it includes diverless mode as a feature.

All this will happen so fast there won't be time to perfect a mixed use car.

By the time they do perfect a local driverless car virtually all cars on the market will already have long-distance driverless mode. And the ones that don't will probably get retrofit kits.

So at no point will mixed long distance ever be a thing - by the time the technology is able to do that, there won't be any cars that need it.

> Would it still be an advanced net if you had humans in the trucks as a 'backup', with the autonomous driving doing 90% of the 'boring' driving down the highways? Humans would still drive the last mile (ha!).

You're exhibiting the AI effect [0]. Could you program the software that makes vehicles drive around "boring" roads? Of course it's advanced.

> I can see self driving vehicles happening a lot sooner than I previously expected. Like 20 years sooner.

Self driving vehicles are already a thing. They're driving around the bay are all the time [1]. They'll be "happening" in the delivery market much sooner than 20 years, like in the next 5. The state of California has already started the regulatory process [2].

[0] https://en.wikipedia.org/wiki/AI_effect#AI_is_whatever_hasn....

[1] https://www.google.com/search?q=self+driving+vehicle+spotted

[2] https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/auto

> You're exhibiting the AI effect

At first I had no idea what you're referring to, but then I realised the confused stemmed from my typo. I meant to ask whether it was a significant advancement, not an advanced net/AI.

I know self driving cars are a "thing" already, but only in very limited and constrained circumstances. Last time I checked, Google's can only drive on roads where they've meticulously already mapped out everything from the height of the curb to the exact location of the traffic lights. That's not to discount their work, but it's still in the very early stages of the self driving dream of having a car that can drive you anywhere while you watch a movie.

Self driving cars technology is AI. It is an advancement, a very significant one that will save many, many lives every day. Self driving cars are not in an early stage. Self driving technology is already worth in the billions. See the GM acquisition of Cruise. These cars have driven hundreds of thousands of miles on real roads that other people are also driving, probably including to a movie theater. You are discount the work and you're doing it from a place of complete ignorance.

Woah. I'm not quite sure where the negative sentiment came from - I'm not trying to discount anyone's work, and there's definitely no ignorance. Maybe I'm not optimistic, but I'm definitely not being ignorant.

> Self driving cars technology is AI. It is an advancement

If you go back up the comment chain that I was replying to, you'll note that I don't dispute (or even talk about) whether self driving cars is AI or not - I don't really care to take part of that conversation.

I was commenting on whether if you put self driving tech in trucks if it 'can only' self-drive on the highways, and require human drivers to drive the final mile, is that a significant enough improvement to make it worthwhile deploying it?

I know we've made a significant amount of progress when it comes to this sort of stuff, but we're so much further away from the dream of ubiquitous self driving cars (or even where human-driven cars are illegal or have their own financial/legal drawbacks). I know (Google's) self driving cars have driven x-thousand miles, which is crazy impressive, but they've only done that on like .1% of US roads[0]. They also have trouble driving in varying conditions, like rain or construction or whatever. I can't see ubiquitous self driving cars for consumers really happening in the next 20-50 years.

I can see, however, this bits and pieces of this tech make its way into cars (like Tesla's autopilot, or the self-parking features) a lot sooner than that. Or into trucks driving down the highway. And then Uber/GM/whoever launches self-driving transport in select cities on select routes, and then all over the city, and then in every main city.

[0] Totally made up number, but I assume you'll get my point.

Add the severe driver shortage that has been taking place in the tucking industry for about 7 years plus and it will happen even sooner.

Well I find it always incredible that the point of self driving cars that is most needed is the hardest part and least likely to occur anytime soon, that is driving in bad to very bad conditions.

though with the prevalence of texting while driving maybe cars will not necessarily need to always drive but simply take the needed emergency action

There is a lot of intercity driving that could be rendered unnecessary with an automated intercity system.

This may also make the shift easier on the current labor pool, as they could migrate to local delivery work.

One accident might set back consumer driving for 50 years, but not commercial. The safety record simply needs to be much better than human drivers, which shouldn't be too hard, because one driver is already taken out of the equation. And there are billions upon billions to be saved, in our corporatocracy of the USA, it WILL happen. Bad roads shouldn't be too much of a problem, they can just use humans on the bad roads, the bulk of our highways are in good shape. Snow is an issue, yes, but remember that they will have the highways perfectly GPS mapped, great reaction time, and commercial vehicles can have big ugly shrouds to keep cameras clean. There is also the possibility of hiring temporary human drivers for the worst snow days, but I don't see that happening because I honestly think the robots can have a much better safety record in snow. The plastic bag/rock thing could be a problem, that could be partially solved by just assuming it's a rock, pulling over to the side and waiting for a remote human to identify it, and clearly mark the truck as autonomous so the humans on the road know to expect different behavior from it. That's not a perfect solution, but we have thousands of the smartest people working on it - there is enough money at stake, it will be solved.

Public and political perception need to consider autonomous commercial trucking to be safe, rather than safer.

"Safer than human drivers" all too easily becomes "we can't control when these robot trucks my crash into a mother driving her kids to school," when you put it into the political regulation arena.

So it's all fine when a lorry rear ends and pancakes a car because the driver fell asleep?

No, but at least from the PoV of public perception the situation is easier to deal with in a way that allays fear.

In your example that one drive is taken off the road. That drive is not a problem, and people assume that the remaining drivers are, for the most part, fine and unlikely to make the same mistake.

In the automated truck example you can't take that one driver off the road because barring equipment failure they are all the same driver and just as likely to make the same mistake.

If one human driver kills in such a manner, people will distrust that one human driver (or at worst that one trucking company, if bad training or other compliance issues are suspected), if one automated truck kills in such a manner people will distrust all automated trucks.

Human beings don't have the ability to just attain knowledge and experience by a software updated, autonomous vehicles do. Yes, every autonomous truck on the road may be controlled by the same software exhibiting the same behaviors, but unlike a human they can all automatically be updated with new behaviors and share experiences with each other.

This is precisely the goal of IBM's Watson if you will remember, an individual doctor only has their own experience - Watson can aggregate data from a virtually infinite number of data points and that knowledge and experience is shared across the network.

Will accidents happen? Yes. But we can learn from them and the collective as a whole benefits, not one individual instance.

I agree. As probably does everyone here.

Now convince the herd that is the general public, who don't really understand the technology, don't have time for a long explanation, and tend to react with instincts and emotions that haven't had much by way of firmware upgrades for quite a few generations.

Likely it will be even better than you describe. Multiple competing companies all with their own software copying what the other teams are doing right. When one fails the others will continue on with lessons learned.

Probably a lot like software development today except with a lower threshold for error, and hopefully the appropriate increase QA.

Nobody is saying this is logical. The point is that people will use this as a hyped-up political foothold to point out why automated drivers are "unsafe" as well as "taking our jobs".

Fun fact: automated elevators were considered to be "unsafe" and "taking our jobs", too. Somehow we got past that one!

And elevators sometimes kill people, yet few people are afraid of them.

Admittedly, deaths are very rare, especially per-trip, 27 per year or 0.00000015% per trip -- self driving cars should be able to get into that level of safety, current automotive fatality rates are around 1 in 100M (or .000001%/mile)


Oh, I absolutely think we will get past this line of thinking. Automated driving is the future. I just don't think it will be a smooth transition.

> political foothold

Driver's union versus transportation industry. I have a hunch which way the politicians will sway.

No, but if a driver falls asleep it's clearly their fault, if they survive the accident they will be charged with manslaughter. If a driverless truck rear ends a car because of a glitch or a mechanical fault where it didn't see the car in front for some reason, then I can only imagine the corporation recalling all of them at huge cost, and possibly paying millions of dollars in damages. I think the point is that when humans crash the responsibility is easy to establish, but we have no framework to work with once computers crash(and they will). And no, "they are still safer than humans" is not a valid defense when someone dies.

The company making the vehicles will be responsible for any damages, and paying a massive insurance premium I'm sure. There will be legal consequences for gross negligence and anyone damaged will be made whole. I don't see how that's any different than the system we have now.

And the emotional appeals cut both ways. How long before the parents of a child killed by a drunk or sleeping truck driver say "Why was he driving anyway? Why wasn't he replaced by a machine that wouldn't drink a six pack and nod off while at the wheel?"

That's partly an effect of being on HN.

In non tech circles - the vast majority of the world - people are pretty famous for measuring risks by perception and not statistics.

The idea of a robotic truck taking away jobs while also being a risk to life? America would take up mass transit before it buys into that system.

Man I can see the headlines and political/ emotional arguments now. Godless SV tech priests rob Americans of the last few jobs they have. 1% and automation, no job left behind.!

It's useful to remember that People refused seat belts until the argument was irrefutable. Hell doctors didn't wash hands regularly until it was drilled into their heads that they were infecting patients.

And when it come to automated personal care - I bet that People will utterly, vociferously, oppose the loss of their agency in controlling their means of movement.

People have romanticized the man and the horse, the man and the bike, the car - but it's because it's a matter of controlling that power.

Selling that to the American public and the world at large is a multi generational effort.

Maybe the American reaction will be like that, but my feeling is that if that happens, then China will take advantage of it by pushing automated driving while ignoring the emotional arguments (authoritarian governments have their advantages in this kind of things...)

And then, the rest of the world will have to follow so as not to be left behind an industrial revolution...

America is not a single unified entity. All it takes is one trucking business to outcompete another for all the trucking companies to follow suit in the following decade.

As for individuals, the government might gives tax breaks or ban the things. I think it is most that likely early adopters will use them and brag to everyone how much they save on fuel and airline tickets and how more (work, reading, video games, etc... ) they can do while travelling. Of course there will be holdouts, but they will be quickly outnumbered.

Eventually so few manual drivers will be on the road that they are the exception and 99% of the accidents will be their fault. Then states will start banning driving manually.

Think Monsanto, not Ford/GM.

And why would I want more places I can work, at least when I drive I am not expected to be working.

Your inability to dictate where and when you work will not dictate my ability choose where and when I will work.

Also Monsanto has nothing to do with anything here, unless you care to explain.

> and paying a massive insurance premium I'm sure

Nah, it'll be much cheaper than what they pay for humans since the bots will be in far fewer accidents and insurance companies actually care about statistics.

I think you might be surprised - companies would be insured against a total recall due to software/hardware failure, which would far exceed the cost of any accident a computer could get into, so I guess the premiums might stay relatively high.

> Companies would be insured against a total recall due to software/hardware failure

I don't think that's true at all. Recalls are a manufactures problem, the local trucking company is just going to have normal auto insurance just as with his human drivers.

When a recall does happen how do the trucking companies keep on trucking?

I mean surely this happens now, how do they solve this problem?

It doesn't happen. For that to happen every trucking manufacture would have have to recall every truck and that's simply never happened. Recalls are generally more along the lines of part Y has an issue, bring the truck in and we'll replace that part.

The same will be true of automated trucking, parts will be recalled, and companies will have multiple different trucking manufactures to choose from so the notion that all the trucks will be recalled just isn't realistic.

I guess airline companies are insured against loss of profit in case their airplanes get recalled and they have to do without. And probably the cost of renting airplanes from other companies so that they can keep on going.

> How long before the parents of a child say ... "Why wasn't he replaced by a machine?"

The ultimate point being that the answer might be "decades." It might take that long for the prevailing opinion amongst the public and/or politicians to change that way. And it is affected by the actual accidents involving automated vehicles in the short term.

Did cars or trains need to be perfectly safe? Even modestly "safer" autonomous vehicles would save thousands of lives every year and provide priceless benefit to society, giving us millions of productive hours and giving people that can't drive (the elderly and handicapped) new independence.

> Public and political perception need to consider autonomous commercial trucking to be safe, rather than safer.

No, safer than humans is the bar.

That should be the bar, because it's the morally correct one. But public opinion is far from rational, and corporate interests would choose the suboptimal route if they could get away with it.

Why do you think a truck driver is more expensive than an autonomous driving system? Have you priced enterprise software lately? It's not like you'll pay a flat-rate one-time fee and have your robotic driver available to you. You'll end up paying based on your vehicle capacity, on the annual mileage, on the annual driving time, possibly even on the market cap of the purchasing company. Then you'll pay annual costs for support of the software.

Beyond a mandate from the government, I have a hard time seeing the trucking industry giving all their profits over to google, tesla, gm, mercedes, and co.

Once the technology is built, it would make sense for the tech company to price their solution somewhere between $0 and whatever the status quo costs. Then they will want to sell as much of the tech as possible, so the pricing will be such that it is prioritized by the trucking companies.

I agree re: the rock/bag, err on the side of caution because any potential delays will be dwarfed by the overall savings of autonomous vehicles and lead to safer driving for other road users.

Wanna know the great thing about humanity? We have almost solved this problem, twice:



Is there any evidence that a technology with billions of dollars in potential would not be used if it was only 99.99% ready? I feel like lots of dangerous stuff has been deployed when it was nowhere near 99% in the past, and I'm not sure why this technology would be any different. People talk about it like it's a lot different, but how safe are regular cars? Nuclear power? Coal power? Honestly I don't even know what the state of the science is on cell phones causing brain cancer anymore because I decided that I didn't want to know.

Why does everyone think we'll be smarter on waiting to deploy driverless cars until they are safe, given our history of doing the exact opposite for the last couple hundred years?

Consider the politics of wiping out the biggest employer in 29/50 states. Can you imagine the backlash that even one death would cause?

Looks like on-the-job deaths for coal miners in the US are around 30 per year. I imagine the backlash for even one death would be relatively minor. Anyways, eliminating the biggest employer isn't necessarily a net negative. Unless (until) these trucks can load/unload themselves, repair themselves, schedule themselves, etc., there's a lot of work still to be done by people.

Hell, I can see ways that automated driving can result in a net increase in jobs. Example: if a truck can drive more hours in a day, it'll do more miles per year, requiring more frequent maintenance, requiring more repair mechanics than we currently have. Another: if you're not limited by drivers needing to sleep, you might be able to do a better job of scheduling deliveries to arrive around the clock, requiring more people to be around around the clock.

I'm not saying either of these WILL happen, I'm just saying that automated trucking doesn't HAVE to be a net-negative thing in terms of jobs.

Yes but those 30 deaths are people knowingly working a job that society understands is dangerous, not a family in a minivan driving to grandma's house that was unlucky enough to be in the wrong part of the highway at the wrong time.

That is a huge difference to most people.

> not a family in a minivan driving to grandma's house.

A total of 3,660 people died in large truck crashes in 2014. Sixteen percent of these deaths were truck occupants, 68 percent were occupants of cars and other passenger vehicles, and 15 percent were pedestrians, bicyclists or motorcyclists. The number of people who died in large truck crashes was 16 percent higher in 2014 than in 2009[1]

One incident would hardly compare to the ~3000 non-trucker deaths ever year due to the trucking industry. I don't think it would be all that tough to find the grandma of one of those 3000 to show up in court in defense of automation. It will be MADD all over again.

[1] http://www.iihs.org/iihs/topics/t/large-trucks/fatalityfacts...

I agree. I was just pointing out it's a fallacy to compare coal miner's deaths to deaths caused by automated trucking. Whatever happens my only prediction is that it will be messy and polarizing :-)

Thanks for bringing data and scale to the conversation...

I rather see it in the following terms: some profits are up for grab, who's best positioned to take them?

I don't live in the US so I might have an inaccurate perception of everyone's leverage. However, truckers seem rather disorganised, compared to superstore chains. If big stores eat up 30% of the savings, and offer the remaining 70% to their customers as lower prices (minus a couple % spent in lobbying of course), I would guess they'd had their way. Ditto for Amazon, Alibaba etc.

Look at Chinese importations: politicians sometimes make big protectionist talks to their constituents, but take no measure effective enough to annoy lobbies. And I think importations represent more jobs than transport.

Company interests generally outweigh individual interests in America. Besides, we live in a global economy, so putting off technology that is profitable (by a huge margin) today only ensures more American job losses as companies can't compete with foreign competitors. I predict that the driverless car revolution, whenever it happens, will be the issue that pushes Universal Basic Income to the forefront of the political landscape.

What Uber did to cabbies, Driverless truck fleets will do to truck drivers

It would definitely be deployed at 99% ready because that would likely mean "safer than people". The quote talks about how we are at 90% but going from 99.99% to 100% for deployment would need a mathematical revolution. Really going from 90% to 99% will also need a revolution. At least on the order of deep learning.

Safer than people is not 99%. The accident rate is around 74 per 100 million miles (and fatalities is 1.13).

It's unclear exactly how to turn that into a percentage, but no matter how you do it it's way way more than 99%.

Say an accident takes 5 minutes, and people drive 30 miles/hour. Then that works out to 99.999% for humans. If you use the numbers for fatalities then it's 99.99999%.

That's not even in the same ballpark as 99%.

The average odds of a human-driven vehicle getting in an accident per 600 miles of driving is about 0.13%, according to the NHTSA. What's crazy to me is that, by the numbers, 99.99% safety would already be 13x better than our current situation. Imagine cutting traffic deaths down by a factor of 13. And that's still not good enough.

I think we need to get over the idea that computers are perfect, especially with regards to how artificial intelligence works. There will always be a situation that is unaccounted for, especially in our non-deterministic universe. It may be uncomfortable to think that a life was taken due to an error in processing, but I feel no better about having a child of mine in a car with a barely experienced friend driving and making many more errors in processing.

I don't know what 99% means in this context, but humans are pretty terrible drivers, so I'd be cool with robot-driven automobiles as soon as possible. If the accident rate is below that of humans then it's ready in my opinion.

Reliable self-driving highway trucks are indeed quite far off. And the OP does at least admit that entry/exit ramps would need to be massively upgraded to support peloton-style semi-automation. Full automation would require a huge investment in road maintenance that many jurisdictions currently skimp on. At some point we'd be far better served to seriously invest in a high-capacity high-speed railroad network. It could move people and freight far more efficiently and could be far more easily automated than automated semis, which are really just a kludge on the current overbuilt and undermaintained highway system. The money follows the infrastructure you have. If we have to rebuild our highway infrastructure to accommodate self-driving trucks maybe we should rethink whether those trucks are the right way to ship goods over long distances after all.

> Industry players are aware that just one serious accident on the roads involving a driverless car could set back their development by decades.

It'd be awfully "convenient" for the truckers & taxi drivers if this were to happen... I predict a dirty tricks campaign with one or two engineered "accidents" to safeguard their industries.

Vehicles, autonomous and current ones, will need incredibly reliable security to prevent remote hacking.

Yes, because computers and software aren't buggy by themselves without human intervention.

It's not a question of whether they're buggy, but whether they are better than human drivers despite the bugs

That isn't the question in the context of my post. I was responding to someone speculating that truck drivers would "engineer" one or two crashes to protect their industry, as if that isn't going to happen anyway.

Sadly, humans are buggy when it comes to assessing risks and their own competence, so autonomous vehicles need to be a huge amount better than human drivers in order to win out.

I can see many road hazards being handled much better by a computer. Depending on what technology is used, fog could have no effect on what the truck can perceive. Also keep in mind that the trucks could talk to one another, notifying each other of road hazards/debris on segments of road.

They were talking about highways exclusively. Using regular drivers to pick up trucks from highway exits and drop them off at entries. Its in the article.

It is likely that fewer cyclists will get crushed like a grape under a semi that's instrumented for self-driving. If it works better, the technology will be launched somewhere in the world, and economic pressure will spread it.

I think the opposite effect would happen and it would be safer for cyclists. There are already cars being equipped with pedestrian and cyclist detection. Sensors can see all around the truck vs the driver view having many deadspots. The computer could be forced to slow down or use safe passing distance while drivers have the free will to break those laws.

Also the computer probably doesn't have a bias against cyclist in the first place.

I'm agreeing with you: Fewer cyclists would die due to automated trucks.

Cyclists tend to avoid freeways.

You can find quite a few videos on YouTube of cyclists on freeways. There is a video from a couple years ago of a bunch of cyclists holding up traffic on the freeway in Austin because they felt like riding on the highway.

I drive in Austin EVERY day. They tend to prefer less-traveled roads. Thanks for the downvote for posting actual facts that I know from first-hand experience.

The article is talking about automation once the truck is on the highway.

"We still need to create on- and off-ramps so human drivers can bring trucks to the freeways where highway autopilot can take over."


Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact