I work with two companies developing true autonomous trucks and the conclusions in this article are dated and not accurate in light of the bigger picture.
1) The model most are working toward is "hub to hub" not point to point. As in, A human driver drives the truck from the warehouse or port to a hub on the highway outside the city surface streets. Then they get out of the truck and pick up another one that needs to go back into the city. The truck then drives itself on the highway to another hub near its destination where another human driver completes the trip. This works in multiple ways - it allows human drivers to deal with the hard part of automation while also avoiding the terrible away from home lifestyle, and it allows you to have dedicated autonomous trucks for on-highway separate from your human-driven fleet (you can swap the trailers instead of using the same cab). You'll see these hubs start cropping up in a couple years.
2) On-highway autonomy is not that far away anymore. It's definitely not a "next year" thing, but mid-late decade is a near certainty based on the current state of the tech. That's a long time for people used to software, but not a long time compared to the lifecycle of a truck.
3) There seems to be an implicit assumption in this write-up that autonomous systems will remain horribly expensive, which isn't true. They're expensive today because each generation of development is effectively custom built in small quantities. Just like Starsky said their camera system would drop in price with scale, a "ready for production" generation of autonomous driving kit in a few years will likely be close to 1/3 the current kit cost, which has very large impacts on the total ROI calculation.
I don't want to make it sound like there aren't big challenges left to solve - there are - but there's also a reason so much investment money has flowed into this space. The ROI does make sense, just not on the original hype-cycle timeline that a lot of people reported.
Trucker with a CS degree here. I'm very skeptical of the feasibility of autonomous highway trucks that can drive outside very tightly defined parameters or ideal situations. That being said, it just occurred to me, how will these trucks react to or let alone detect tire blowouts or break failures/fires on the trailers?
Edit: I know from experience that getting a project to what we think is 95% there is relatively easy compared to the last 5% that make the product/project 100% viable. The real life testing will kill you if the edge cases don't. (I don't mean that literally, although...)
CS came first. At the end I had a small internet business at home that chained me to the computers. An alarm about a defective RAID just as I was starting a short vacation triggered the career change. I couldn't see myself go back into a cubicle and I always liked to drive so I decided on the spot to become a trucker. So I sold the business, took a course, pass the license and here I am 15 years later, 100 pounds lighter because I'm no longer behind a desk getting fat. I'm doing local and regional deliveries with a 53ft and have a 20km bike commute. I probably make 1/3 of what a computer job would pay but to me money is not everything. To tell the whole truth, I was simply burned out. I just came back to programming on personal projects just a couple of years ago and I love it.
As for comparing industries, I don't know, how can you compare when one has really small margins, has no technology besides tracking and logistics, is really change averse and has basically an un-educated cheap labor force and the other has hopefully good margins, a well paid skilled labor force and creates technology.
One thing for sure, a good or a bad boss is the same in both industries lol.
Yeah, and what about theft? If the truck is programmed to stop if there's a person in front (one would hope so), then how can you stop thieves lifting the whole thing?
How is that prevented now? Also, a relatively common form of theft now is the human driver absconding with the trailer. Also, a self driving truck would presumably be generating high resolution 3d scans of any potential hijacker.
people are generally a lot less brazen towards other humans compared to how they interact with automations or things that are remotely surveilled rather than with a human attendant.
trains with unarmed guards generally don't get robbed.
> The model most are working toward is "hub to hub" not point to point. As in, A human driver drives the truck from the warehouse or port to a hub on the highway outside the city surface streets. Then they get out of the truck and pick up another one that needs to go back into the city.
What's the competitive advantage of this setup over a train?
The thing about trains is that they gain cost efficiency in exchange for a massive time penalty. In most cases, you have to truck your goods from a warehouse to a rail center, unload them from the trailer to a rail car, wait for the entire train to be loaded and prepped, wait for a slot on the rail network, travel at normally ~45mph, and then do the reverse at the destination rail center. A trucking hub can swap a trailer in maybe 10 minutes and there's no need to group shipments together - you can send a single trailer load of stuff from point A to point B with minimal interruptions just like with a normal truck today.
I agree that's a concern from an investor's perspective. But:
1. You benefit from the massive subsidy America provides for road transport
2. You can serve places that don't have good rail connections
3. If you're moving standard trailers, your 'hubs' don't need costly build-out like cranes. Your MVP hub is just a car park - or maybe even just a quiet side street. So you can serve a town that needs 2 trucks/week, instead of only being able to serve towns that need 40 trucks/week.
Not my area but seems like if you swab cabs you would be able to keep your fleet either on the road or charging & getting maintenance.
> The charge time offers time for maintenance & switching trailers
> Charge times keep dropping & miles per charge keep increasing
> Batteries currently make up almost half the price of a car. Battery prices are dropping though. Not sure about cab prices. Also these cabs might be very cheap without human comforts in comparison to the battery, so if battery ends up being bulk of cost, you might gain little value there.
Anyways just 2 random cents from someone who finds interest in this but doesn't actually know anything.
I think this point is key "Those margins would jump to 58% if that remote driver only needed to pay attention for the first and last miles." - if you're going for #1 you still have a human in the loop for the first and last mile. If I understand the original post correctly (and I may not be), going from human in the loop at beginning and end-of-trip to no human in the loop at all is where the ROI vanishes
"While this is a lower margin business than software’s traditional 90%, we expected to be able to get to a 50% margin in time.
It took me way too long to realize that VCs would rather a $1b business with a 90% margin than a $5b business with a 50% margin, even if capital requirements and growth were the same.
And growth would be the same. The biggest limiter of autonomous deployments isn’t sales, it’s safety."
In either case, returns are gated on fleet deployments, and the teleoperator approach seems more likely to deploy sooner at scale. Is it that VC's in this case were not investing in logistics, but in AI?
You're treating VC's as rational economic entities with only money at play. VCs have personal interest, and sometimes have other stake in the situation. If your background is coal mining, and you are a VC who also holds significant stake in coal mines, you are very likely to fund CCS ventures with high risk for reasons which do not equate directly to the ROI inherent in that specific CCS venture and its future share worth.
See Twiggy Forrest and Mike Cannon Brookes. They're investing in solar to power a long-line HVDC feed from NT Australia to Singapore, and Forrest is funding H2 plants in Gladstone, Qld Australia to supply a future Hydrogen market which doesn't yet exist. Why? because they WANT TO.
"oh, if you want to, then you aren't a VC" .. sorry, its not always just about the money.
You cannot say all things are equal, then x5 the total revenue of a company with significantly lower margins and say you’re comparing apples to apples.
Like to like comparison would be six companies each making one billion of which half have a 90% margin, other half having a 50% margin. If you only have the capital to invest in three, which would you pick, assuming all other factors were identical?
Capital requirements in this context are the funds required to reach profit. If both companies are the same except for profit margins, the 90% margin company would reach profits sooner that the 50% margin company. In my opinion, odds they would have the same capital requirements to me sounds like magical thinking.
That said, even if the 50% margin reaches 5 billion in revenue, in theory cash flow might be more that the one billion in revenue company over a long period, but VC invest to flip investments, not for cash flow.
Well, the company in the article looks like something that would be profitable starting from a very small fleet. I can easily imagine it being profitable with 4 or 5 trucks.
And from the numbers on the article, it would get a 50% margin, with something around 3 to 5 times ROI in an year.
That's certainly not young Google, but why the hell would somebody refuse to finance this pivot? (There's something clearly missing, either from the company of from the VC's side, but VCs seem to get on that situation again and again.)
Those were not real numbers and were based on set of assumptions that did not hold true, including but not limited to the core assumption — Starsky was not able to reach an AI drive autonomy level needed. They also assumed margins would not be razor thin if two or more logistics companies succeed in creating autonomous trucks; governments would not allow this, it’s a pipe dream.
> Starsky was not able to reach an AI drive autonomy level needed.
The numbers were for fully manual remote driving. This thread is all about asking why they closed down instead of going in the fully manual remote option.
The article makes it look like the manual remote driving was perfectly viable already. Although there are obvious connectivity problems for scaling it over huge distances.
> They also assumed margins would not be razor thin if two or more logistics companies succeed in creating autonomous trucks
Any midsized investor would expect to reap the high margins during the transition, and deal with (maybe sell) a fully established very large and lucrative logistics company afterwards. A smaller investor would expect to sell it as a service to all logistics companies, and reap a share of that 3x ROI with much smaller investment during the transition, and then probably sell the company for relatively cheap.
Not able to find any proof Starsky was successful in legally and remotely operating trucks at speed at any distance on public roads without special constraints that would not representative of normal daily driving conditions.
For example, this test was on a public road, but it was closed to the public during the test:
That math doesn't add up. If growth were truly the same many would prefer the more than 2x gross profit dollars. However in many cases the growth profile is different between such different models
That's really sad, because it's a good idea, trucking has a massive labor crisis for all the reasons they detailed (only paid for hauling, not waiting, combined with the race to the bottom in transport pricing where independent truckers and trucking firms both are cutting their own throats to win contracts).
Yeah, the broking model is responsible for the trucking shortage. You start paying for trucker time at origin/destination, and the supply chain would become a whole lot more efficient, and trucking wouldn't be (as) terrible a job.
Hmm I wish people would stop saying AI is "just pattern matching". I think it's pretty clear that a) state of the art LLMs have gone well beyond basic pattern matching and b) there's no fundamental reason (that anyone knows of) to believe that humans aren't "just pattern matching".
I mean I don't think we'll have fully autonomous vehicles using AI for a very long time because it basically requires strong AI... But it also doesn't seem like research is especially stalled.
I work for a Mining company that has autonomous dumptrucks (among other things), and the autonomy has proven its ROI quite nicely. So much so that we are expanding its use and automating more things like light vehicles and drill rigs etc.
Its not cut and dry / black and white, there has been some loss in efficiency in some areas but that is a learning process and we're constantly reviewing and updating.
For us its about economy of scale, and given we control our mine sites 100% its a bit simpler than fwy's and hwy's etc.. we dont get inattentive drivers and drunk drivers and teenagers in blinged up ricerboxes.
The value its added has shown me that autonomy on the public roads is definitely going to be a thing and probably very soon.
I would expect insurance companies to be the biggest drivers... offering discounts to vehicles that are not driven by humans. follow by municipal councils, who will notice that the more autonomous vehicles there are, the fewer accidents there will be and the better traffic flow is thereby reducing the need for road expansions and traffic management upgrades.
This is the case for all transit, known as induced demand. Increase the number of lanes on a road and you'll get as bad if not worse traffic than before as the increased capacity results in more people driving.
This is a fantastic read. But don't agree with the conclusion. It's myopic.
Few problems I have with this,
- The cost saving of 600K/truck/year is low relative to the cost saving with lower autonomy. When improving efficiency in any operations system, the ROI keeps getting lower and lower with further improvements. Does not mean that the investment is not worth doing
- The efficiency principle of software relies on scale. There's no reason to build autonomous vehicles at all if you own a single truck but this makes sense when you have a large fleet. That's why software earns with scale. 600K/yr is a single truck saving. It becomes significant over the scale of trucks and the years.
The unit economics is wonderful to go through. Appreciate the research!
> The hardest 1% of the technical problem, automating the surface streets and interchanges, would end up being worth only about $600/truck/yr. Level 4 truck autonomy has less value than a daily coffee.
Note that savings is $600, not $600,000. For a $100K truck, that’s 0.6% per year. Not nothing, but they’re happy to let someone else do the Herculean effort to get to L4.
For each programmer on a team that could do it in a year for, say, $240K in comp, you’d need 400 truck-years ($40M worth of trucks for 1 year) to make it a break even proposition. How many programmer-years do you think it would take to build L4 autonomous trucks? How many programmer-years have FAANG alone (leave out auto manufacturers or taxi apps) spent over the last decade on autonomous driving?
Autonomous trucks are just some minor differences from Autonomous cars. No auto manufacture dares risk someone else come out with an autonomous car that is enough safer than human drivers that governments mandate it. This is mostly about all the patents needed to make it work that the first mover will get. As such all car companies are investing in this to ensure if someone makes it works they will have it too, or at least enough useful patents that they can get a good licensing deal.
I think Tesla is the only company attempting this alone, all the other car manufactures are just in some form of partnership with others. These partnerships are known to include the big truck manufactures. I suppose there are a few car companies not investing in this (either tiny, or not selling cars in western countries).
Great article, but I think the conclusion is a bit off.
We also need more research into autonomous driving, to be able to drive trucks safely on highways. And automating that part you see as highly profitable.
So, if anything, the article shows how important more research into this topic is, especially if we would factor in the two disproved assumptions, you mention in the last section.
Second, I think it is a bit naive to think that truck manufacturers/tech companies will just sell trucks for a similar fix price to traditional trucking companies.
I think they will either
- lease them at an expensive rate (getting most of the additional margin, since they also deliver the value)
- offer them as a service, such that trucking companies only manage contracts (also expensive)
- found their own new companies or buy trucking companies to handle the contracts
One thing that seems off to me about autonomous trucking is the following:
Imagine a chain of autonomous trucks on a highway — isn't this basically a less energy efficient, less capacity version of a train that pushes a part of the costs onto public infrastructure?
Granted, trucks can also cover the last and the first mile(s) with much more flexibility, but a well designed system that allows you to quickly move containers from trains to trucks and vice versa could do just that as well.
Also: while this might not be as relevant within the US, on the rest of Earth truckers often also handle the customs at national borders and need to interact with certain laws. E.g. last week Bavaria, Germany banned trucks from taking non-highway routes to Austria. Not sure how an autonomous vehicle would honor this, or how police would be able to stop them. The mechanisms for doing all the things beyond just driving are just not there.
I think a big reason why trains aren’t used more (and why highway trucking is booming) is because the industry is sclerotic. In the US at least, it is one of the most conservative industries there is at all levels: regulators, companies, and workers. And because of the difficulty of switching, it’s hard to incrementally innovate on the system.
That said, there ARE companies innovating in this space. This seems like an obvious improvement on the status quo: https://moveparallel.com/
With battery-electric systems (enabled by lithium-ion), there is no massive benefit from centralizing motive power in the locomotives (thermal engines gain a lot of efficiency by scaling up, electric systems can operate efficiently at any scale), so you can just have individual packet-switching-like rail cars. Containerization systems (such as cranes, etc) open up the trade space for switching payloads without requiring complex and land-intensive switching yards for everything.
And there should be more room on the freight rail system as we decarbonize. Coal was 25% of rail volume in 2009 (much higher, if you go by total originated rail tonnage). There will be room on the railroads for more volume.
That isn't actually clear. While trains claim they are the most efficient, they use measures like ton-mile that make them look much better. Trucks are often used for light things which means the ton-mile would go down. Trucks are often used for small loads, trains to all those different destinations with small loads would be less efficient. Sure steel wheels have less rolling resistance than rubber, but it isn't that much less. Long trains are more aerodynamic, but trucks already have a fair amount of length to spread that would over.
As such it isn't clear how much more efficient trains would be in the real world of freight.
Note that the above is about freight. Transporting humans as a number of different factors making the above inapplicable.
Go ahead. I am interested in the numbers that would support your theory. My k owledge is that trains are by at leasy a magnitude more efficie t than trucks.
> Imagine a chain of autonomous trucks on a highway — isn't this basically a less energy efficient, less capacity version of a train ...
In principle, yes. And a train is basically a less energy efficient version of a barge or a ship. The trade-off lies obviously elsewhere (and is complex). Typical problems are loading and unloading goods (which is often quite slow, even when containers are used) and re-assembling trains en route when wagons have different destinations. B2B in bulk is were trains excell, such as deliviering coal from a mine to a power plant.
> ... that pushes a part of the costs onto public infrastructure
Only if trucks are not adequately charged via taxes and tolls. To get foreign trucks involved in financing the road network, many European countries have moved from pure tax-based systems (on petrol and the vehicle itself) to toll systems in recent years.
> Imagine a chain of autonomous trucks on a highway — isn't this basically a less energy efficient, less capacity version of a train that pushes a part of the costs onto public infrastructure?
>- your infrastructuur is re-usable by the general public.
It's the general public's infrastructure to begin with and for most countries this is also true for rails and their use.
Trucks damage the surface a lot lot more than cars which currently in most countries isn't really accounted for.
If it was (trough toll fees notably higher for trucks for example tho imo tolls are annoying and generally badly implemented) then it becomes a downside.
Intermodal logistics systems already use trains for quite a bit of the truly long distance travel in the US. (Probably less so in other areas.)
To your other point, yes, people are still needed for various purposes in the transport process. And that's sort of the issue. If at the end of the day, you need people available at various points, trying to automate past maybe some of the long haul highway sections (which probably has safety etc. benefits as well) doesn't really buy you a lot. Elaborate schemes of depots at highway exits and so forth have clearly diminishing returns.
You have similar but different considerations for automobiles. Unless you're truly going to have door to door in at least most conditions autonomy, automating long boring highway driving is really the big win even if it doesn't give urbanites in particular the personal chauffeur they want.
> Not sure how an autonomous vehicle would honor this
What? It's not like an autonomous vehicle is a 100% black box that you tell "go to point X" and everything after that it figures out itself. Of course it's based on traditional route planning and you can just tell that route planning to not consider banned roads.
I used to own an OTR trucking company. The biggest issue for us other than finding drivers like the article mentions were the FMSCA regulations around the duration a driver is permitted to drive on a given day. We would be forced to stop a few miles short of a dropoff location because the driver ran out of driving time and was forced to enter the sleeping berth to rest. Companies like Amazon solve for this by driving in teams of two per truck so you can swap drivers, but finding even a single driver for us was a challenge. I'm curious what the current state of regulations say for autonomous trucks. If they are not capped on drive time then that alone would be a huge win for companies.
We are incredibly far away from autonomous trucks. I am not aware of a single truck that you can buy that offers assisted (as in you press a button) lane switching. There are lane keeping assistants and assistants which use cameras to check whether switching lanes is unsafe due to other vehicles on the road.
Waymo seems to be working on it: https://www.youtube.com/watch?v=Z09jND0G6HQ but the fact that this 20 second maneuver is warranting a video in their eyes should tell you how complicated it is.
It seems like AI truck driving, at least hub-to-hub, is already pretty well solved. Here's a test example (in Germany) that incorporates electric trucks which only need enough onboard energy to deliver "the last mile" : https://www.youtube.com/watch?v=_3P_S7pL7Yg
I believe the conclusion that ROI is poor is misleading:
Imagine a new truck company created by one or more tech giants.
A) Using state of the art tech.
B) Assuming mostly full autonomy.
C) Factor in the cost savings.
Conclusion is this company would very largely outperform traditional trucking companies in terms of profitability. And it would be extremely scalable.
If that holds then it follows that autonomy is simply the last component necessary to be able to disrupt the trucking industry as is.
The real question is to what degree autonomy is needed to be able to start doing this.
$200,000 setup but $65,000/yr saved, before counting efficiency. Adding in that drivers can only drive 11/24 hours (legally), you get $130,000/yr saved and drop your initial capital outlay in half.
Automated trucks pay for themselves in just a few years — if not immediately, due to cutting your fleet in half.
In practice, the fixed costs of R&D in the AV space are pretty large. If for example it takes 500 engineers that cost 250k per year to make the trucks self driving, then you're spending 125M per year just in engineer salaries. That's of course excluding any of the costs spent on things like data storage (cameras and lidars generate a ton of data), compute (ML eats a ton of processing), data labeling, mapping, vehicle testing, fleet management, regulatory compliance, etc.
There are 2M semi-trucks in the US, so if half are automated (1M), you can spend $1B per year for $1k per truck per year. Which doesn’t substantially change my calculations. That’s my point: the unit costs are so ridiculously good that pulling $10k/yr in tech overhead, or $10B/yr total, you’d still come out ahead in 5 years.
Phrased another way: we can spend $100B across a decade to develop automated trucking, and the consequence would be our automated trucks break even in three to five years instead of two to three years, after amortizing the R+D.
I didn’t assess the operations because I couldn’t find a good metric on how much that costs now - and hand waved fleet maintenance, testing, regulation, etc as approximately the cost of the current burdens for those things.
Oh, I agree with you, the numbers are mind boggling. Just in the US, there are 3 trillion miles driven per year, so even capturing a small part of that (be it through automating long haul trucking, goods delivery, or robotaxi applications) and getting any kind of per mile/per hour revenue makes some pretty large numbers.
It's just that the AV players absolutely need to operate at scale, as their fixed costs are really high compared to a traditional trucking company.
I’m not sure that R+D needs to be at the operator versus manufacturer level. For instance, manufacturers not operators bear the R+D cost of better diesel engines.
You can retain (relatively) small fleets if the manufacturer is amortizing the cost of technology across vehicles sold.
It really depends on whether the AV companies decide to focus purely on manufacturing and developing AVs as a product (profit at the point of sale), offer them as a service (similar to "power by the hour" in aerospace), or decide to become robotaxi/truck fleet operators to capture more profit.
You're assuming that L4 is possible and that this incredible technology would be available for merely $200k per truck. I can't fault your optimism, but I think this might be a little out of step with reality.
This isn't just about the cost of a rack of GPUs. I think you're severely underestimating the difficulty of self-driving, it's still not certain that we'll safely and reliably make autonomous, self-driving cars and trucks. And the cost in R&D to make such a breakthrough, assuming it is possible, is a complete unknown.
Slapping an extra $50k on the sticker price of an average truck (which is what I think we both googled to discover it's $150k) and doing some rudimentary maths doesn't result in an answer to the question of whether self-driving trucks are possible, affordable or practical. It just tells you whether some finger-in-the-air numbers you selected sum to >0.
Seems like everyone have completely forgotten everything else a trucker does except driving the damn truck on the highway. They assist with loading and unloading the goods, make sure the goods are adequately stacked and strapped and obeying laws and regulations (at least in my country truckers must have a bunch of certificates to be allowed to transport dangerous goods), handle paper work (this is probably the easiest problem to solve) and probably a lot of other stuff I know nothing about. In addition to automating the driving, you must offload these activities to warehouse workers and others who ship/receive goods.
The autonomous truck also have to be able to handle all kinds of tight and busy city streets, badly thought out warehouse terminals and a million other things that is an annoyance for a trucker but a huge problem to automate.
I had a mental model of trucks going from waiting lot to waiting lot on the freeway, and going the last couple miles either by switching cabs or having local drivers take over.
I think you reduce a lot of the challenge with freeway-only driving between large staging lots adjacent to industrial areas and allowing humans to do the “hard” part of the first + final mile.
In fact, the NY State Throughway already has something like this, but for a different purpose: Combining and separating double-bottom trucks (which are allowed on the Throughway, but not on many local roads around it).
> They assist with loading and unloading the goods, make sure the goods are adequately stacked and strapped
This seems like a problem already solved in the airline industry.
* Pilots don't load and unload the planes. Baggage handlers and gate agents do that. Flight attendants ensure that self-loading cargo is adequately strapped.
* Airline dispatchers (a job which requires an FAA certification) handle much of the flight planning, weather monitoring and other paperwork.
In other words, many of the "other things" a truck driver does are outsourced in the airline context to other employees who are not controlling the airplane.
I don't really see why the DoT couldn't create an analogous "truck dispatcher" position for OTR trucking, who handles all or most of the things you described.
One long-term consequence is that it would further entrench the scale effects of volume shippers (like ports and Amazon warehouses): A dispatcher at the origin would be the responsible person to verify that loads are correctly stacked and strapped, and shippers below a certain scale would not be able to fully use a full-time person in that role.
The difference is that we can look at other R+D of technologies (which cost less than that), the rate of autonomous piloting in drones and cars, and have reasonable idea that $100B-$1T over a decade will allow us to reach the level of freeway driving for trucks.
those tractors don't have to avoid other vehicles or pedestrians, they don't have to obey traffic laws, they don't have to be on the lookout for road maintenance, closures, detours, etc.
(p.s. not an endorsement, I'm personally not a fan of Uber due to their tendency to play the role of Bad-Actor against both their gig-economy workers and customers, even seeking out and leveraging political corruption.
Really low standards for ethical code of conduct.)
But I always consider at least 3 or maybe even more players. Each invest huge amount of money. Now there is oversupply. They must get at least some of this back. Thus they start to compete with prices. Eventually they end up with pretty slim profit margins.
The one thing I don't understand is: "Why don't we make all trucking semi-local?"
It would seem with modern automation that you could place relay stations at roughly 250 mile spacings so a driver can pick up a trailer near home, drive 4 hours out, drop off trailer, pick up new one, drive 4 hours back home, drop off trailer, and go home.
This seems like a much easier nut to crack than autonomous driving. Real estate in a lot of flyover country in America is really cheap. And it has other advantages like gasoline and service are at specified points. Basically, treat trailers like IP packets.
What am I missing? (I suspect some sort of CapEx/OpEx accounting tomfoolery that means everybody wants everybody else to take the risk of actually owning things.)
You're missing the points of failure and the complexity of dealing with them. If the truck supposed to be servicing one leg of the journey has a breakdown, or the driver goes on sick leave, suddenly every container that was to pass via that leg is now stuck. Instead of having 1 spare long haul truck that can do the full journey, you have to have a spare truck per leg (Or you have to send your redundant truck on a long-haul drive to replace the broken truck, and then have them work overtime to catch up with the backlog of containers)
You also lose throughput by introducing potential handover delays. You end up with a bloated admin that is continuously having to fix a shifting constraint as certain legs become bottlenecks, etc.
Part of the justification for the insane expense of the interstate highway system (besides the dubious defense benefits) was that it would act as a check on usurious rates by the railroads. Yes, we could just regulate or nationalize the railroads (that the public largely created via huge land giveaways and other benefits) but it's more American to counter the flaws of one big business welfare program with another: subsidizing auto infrastructure as a giveaway to Detroit and the oil companies.
Trucking isn't always symmetrical. If your current drivers are mostly doing City A to city B to city C, then you need a lot more drivers and half of the time they run empty going back and forth.
On an emotional level, I certainly sympathize strongly with this idea.
I can see how tech often helps widen the gap between the haves and have nots.
At the same time, tech advancements have often improved humanity's standard of living - providing clean water, lighting, refrigeration, cooking, etc.
The basic thesis is while tech makes some jobs obsolete, but they will be replaced by newer jobs, and everyone benefits from the better, cheaper world created by the tech advancement(s).
The main critique for this notion is asking how many of these new jobs are really created, and whether they offer the same stability and benefits.
If this thesis is valid however, the other big issue that remains is the widening wealth gap. The solution for that is probably not banning automation - but rather progressive taxation based on one's wealth. I know some people find this idea evil, but it's a fantastic solution to the issue of extreme inequality.
Tech makes jobs obsolete, but the new jobs introduced in the last few hundred years only add up to a small part of the overall workforce (programmers excluded). CGP Grey went in depth on human replacement automation, and almost a decade later it still holds up: https://www.youtube.com/watch?v=7Pq-S557XQU
The bottom line is that automating a job is obviously a net positive because it frees up a human to do something else elsewhere, while still providing the same added value as before. There will always be something for that human to do, even if not a traditional style job.
The gap, as you point out, is making sure that newly created value gets taxed and provided to that freed up person as UBI, so that they can continue to live and consume those products that the machine makes in the first place, otherwise there is no demand for them and capitalism doesn't work. There is no point in having an army of robots making hats if there's nobody to wear them.
Labor doesn't want to be laboring. It's better in the long term to remove the need for labor than to have people trapped in labor but "ally" with them. We get to enjoy being programmers instead of farm laborers because previous generations succeeded in using capital to win against labor.
What do you expect to happen to these laborers? Will they get to retire in comfort because our society is so rich?
Heck, no. America's philosophy is if you don't work, you don't eat.
The people investing the capital to replace the workers are doing this so they can capture the money that used to go to the workers. And that's what will continue to happen.
You may be assuming that the ones benefiting from automation are happy to live in a world where they are hated by almost everyone, and at best kept safe by using some private army. Imagine living in a world where "eat the rich" is not a fringe idea but a campaign promise...
I would argue the rich would much prefer to let everyone live at some <low but "okay"> standard of living than have a few more zeros on their bank totals and live in constant fear.
I think the world shown in the movie Elysium portrays this reality and its possible outcomes better than I can do. Not the be the best movie ever, but it got this part pretty well in my mind...
> I would argue the rich would much prefer to let everyone live at some <low but "okay"> standard of living than have a few more zeros on their bank totals and live in constant fear.
I suspect many of the rich don't feel they are yet rich enough for this reasoning to be applicable to them.... and each year as they get richer, their subconscious goalposts move a little further.
At some point, obsolete workers have to stop working. Do you want to phase them out slowly at retirement age or something? Keeping them forever is worse for their more numerous successors.
Great and detailed article. I agree with the comment from balaselvam though. On the way towards full autonomy (whenever that might be achieved) you will get there incrementally (like Tesla is trying to do, their constant delays/promises notwithstanding). At some point highway driving will be ok, but even for highway driving it is absolutely imperative to have a very good understanding of how to drive, the level of autonomy needs to be quite advanced. Otherwise you can quickly end up in a situation where unforeseen traffic guidance, debris, loss of communication will result in a situation where the truck does not react well enough.
I doubt it is really possible to design a system for 'just' highway driving of such high quality that no intervention is required within the reaction time of a remote operator without actually trying to fully solve it. This could be limited to highways of course, but a loss of connection in city regions would then automatically mean the truck would have to stop where it is (in the middle of the road with hazards on)
I do like the remote operator scenario for start and end quite a lot though as it would ease the incremental inclusion of autonomy into the truck economy and offer higher safety with machines this big.
Sounds like trucking companies should build a network of well located cheap rooming houses for overnight stays for truckers and scrap overnight cabs. They could even put them next to fuel stops and open the eateries to the public. They could be called truck stops.
With all the hoopla that went into self-driving cars, why do we still not have self-driving trains? Operating on a fixed track reduces the problem space quite a bit, no? So if the technology can't get there for trains, then it will never get there for cars.
The problem is automatizing the existing infrastructure, 'cause it requires great effort (so, money and inconveniences).
I don't think however that the lessons there are applicable to cars as well, and while I think we could and should invest a lot on train systems, the final goal for cars should be reducing them, not replacing them with autonomous systems.
And that one driver now carries all the responsibility. To automate them out, you need 100% accuracy which closes in on impossible to achieve. Any lack of perfection would be down to the automator. So it's a very scary piece to take on when there's no human in the loop left to hold the bag when something goes wrong.
Computers are more accurate than humans in a number of the things that matter for trains.
As the other poster said, the main need for humans is to fix mechanical problems. If the door jams, unjam it (why can't passengers use a different door or figure out how to unjam it>) If someone is having a heart attack do CPR until medical crew arrives, then stop the train until the medical crew is off (central dispatch can teach anyone CPR over the phone and stop the train). If someone is attacked - stop the the attack (unless you are a trained police office you will probably make the situation worse trying to stop an attack)
In short there actually isn't that much that a human can do that a computer cannot do better.
> In short there actually isn't that much that a human can do that a computer cannot do better.
Passengers fixing their own doors and responding to their own emergency alarms doesn't fit most people's definitions of "better". Hence why even though train automation has been possible for decades, most trains and especially most trains travelling long distances aren't.
And in non closed systems the set of problems that prevent fully autonomous cars is the same set of problems that prevens fully autonomous trains, ie, recognizing shit getting into the lane of travel but not panic braking for random trash.
If there is something on the track the only option is hitting it. You cannot switch track (even if there is a switch on the tracks, it can't safely operate in time) You can hit the brakes, but your braking distance is so long that you won't be able to slow down much before you hit it.
I have talked to somebody in the SBB R&D department recently about this and the problem is actually that there has to be at least one person on the train.
Imagine a door gets jammed or there are some problematic passengers. If there is someone on board they can fix the problem in a couple minutes. But on a completely automated train there is no one on board to fix the problem. One solution would be that you send out crews to arrive at smaller stations and fix the issue, and larger stations could have dedicated people, but that is already more investment and less efficient than just having someone on board.
Retrofitting automation for a train network in a wealthy country like Switzerland wouldn’t be a big problem, but it just makes more sense to leave the driver in there. Accordingly, SBB is investing much more into driver augmentation to help the drivers drive as efficiently, safely, and consistently as possible.
Railway is a different problem space than road vehicles. Trains run on fixed tracks, has long braking distance, are much heavier, and accidents can be much more devastating. While self-driving cars may use cameras, lidar and radar for detection and brake on sight, trains need to know exactly how far they are allowed to go and what speed restrictions are ahead in order to brake in time.
Traditional lineside signals and signs do give enough information to drive a train by a computer, but the rail industry has already chosen decades ago to integrate signalling and automation with the infrastructure, which is a completely different approach than today's self-driving cars. As a result, heavy rail and metro trains will not be "self-driving" in the same way as self-driving cars.
Now, if you say trams that run on streets, then possibly. Trams are often driven by sight and has stronger braking so it is possible to apply the same principles from self-driving cars. In fact, Siemens has been developing and testing one such tram. [1] Though I think the problem space is very similar to self-driving cars. There is no steering, but it still has to handle pedestrians and other road traffic.
They exist, many new lines or expansions are highly automated. Retrofits are hard for two sets of reasons:
Political: union strikes (meaning if you want to automate a line you may have to accept it being closed for the duration) and public outcry over a reduction in jobs.
Cost: You often only get the benefit of the simpler domain if you can implement the whole system from scratch, if you have to deal with the existing signalling, engines, (and human-driven trains) that's a complex problem, with massive capital outlay.
> Political: union strikes (meaning if you want to automate a line you may have to accept it being closed for the duration) and public outcry over a reduction in jobs.
For local public transit subway (ny mta, path ny nj), I think the cost is worth it if it means fewer delays.
Talk to your local politicians and demand it. This means changes in state laws as well to be less union friendly in cases where automation is replacing their jobs (I'm not saying that all union friendly laws need to go, but some specific ones)
The problem of automating trains is easier, but train labor requirements are already very low compared to truck requirements, so there isn't much of a win there. A typical freight train has a 2-person crew carrying 200+ TEU containers.
Even if it is technically solvable, don't underestimate the work to be done to comply with regulations. Some might even not be available yet, especially when talking about automatization with AI.
Complex non digital signals. My neighbor is a train conductor and he had an accident because he missed a signal. He almost died and had to be cut out of the cabin. Some passengers were also hurt. Imagine the outcry when a ai train would do that.
Other than that isn't there in London a autonomous train? I remember riding that in 2005 or so
The DLR is semi-autonomous. It has the ability to act autonomously, but it must be started with a button and has an attendant on-board that also fulfills other duties and can be there in the case of an incident.
According to this article (https://railinsider.co.uk/2020/10/27/making-london-undergrou...), it occasionally has to even fall back to effectively level 2 operation (i.e. operator at the front of the train) because increased passenger usage makes level 3 operation as originally built (no platform edge doors or obstacle detection systems) unsafe at certain stations during peak times.
Automated trains wouldn't do that because we can design signals so that they cannot be missed (that is if you don't have positive go/no-go you stop in time)
Automated trains do have difficulties, but they are very different from the ones you see with human drivers.
I wonder how latency is handled. I wonder how close nearby you'd still want the remote driver since someone across the world would have at least hundreds of milliseconds in delay communicating back and forth
Also, what happens in periods of disconnectivity? It seems like you'd still need some autonomous capability then, although it might be lessened to only needing to pull over?
For Starsky's engineering / demo purposes - they could limit themselves to a few safe routes with good connectivity. Just use a couple of cell modems from different networks, maybe add a satellite backup to be extra safe and you should be safe enough if the truck can do basic collision avoidance and lane-keeping.
For the future, the good news is that the world is already headed towards a state where roads come with high-speed connectivity infrastructure for vehicle-to-vehicle, vehicle-to-infrastructure and vehicle-to-x communication. Building reliable, high-speed connectivity to the road makes a lot of sense if you want autonomy to really take off.
This is pretty simple, there’s really no need to give direct steering/throttle control. As an option, it’s good to have, but any real system will default to higher level commands, e.g. drive to this location, pull over, etc.
If they have to have an office in every state that is no problem. Except maybe Alaska there are enough cities that you can open an office in every state very quickly and hire people to live there. In fact work from home might even be useful. I could well see them hiring remote people with a "must live within 100 miles of X small town and have a reliable car". 95% of the time they will work from home, but that other 5% they use their car to drive to wherever the stuck truck is and figure out how to get out of trouble.
> The truck then drives itself on the highway to another hub near its destination where another human driver completes the trip. This works in multiple ways - it allows human drivers to deal with the hard part of automation while also avoiding the terrible away from home lifestyle, and it allows you to have dedicated autonomous trucks for on-highway separate from your human-driven fleet (you can swap the trailers instead of using the same cab). You'll see these hubs start cropping up in a couple years.
That makes sense, but just sounds like a worse version of freight trains to me.
I so hoped this would be about HPs purchase of Autonomy:
Autonomy was acquired by Hewlett-Packard (HP) in October 2011. The deal valued Autonomy at $11.7 billion (£7.4 billion). Within a year, HP had written off $8.8 billion of Autonomy's value. HP claimed this resulted from "serious accounting improprieties" and "outright misrepresentations" by the previous management.
Well written and very detailed. However, it's only focused on the effects of autonomy on currently operating manned fleets. The real question is what a new business will look like surrounding autonomous trucks. Autonomous vehicles aren't built to convert businesses - they're built for safety, convenience, and for businesses we haven't seen yet.
> Mining giant Rio Tinto is running pits at its Yandicoogina and Nammuldi mine sites, with workers controlling the driverless trucks largely from an operations centre in Perth, 1,200 kilometres away.
> Josh Bennett manages the mining operations at Yandicoogina mine north west of Newman and is closely involved with running 22 driverless trucks on the site.
That was then and much has happened since.
> In 2018, each truck was estimated to have operated on average 700 hours more than conventional haul trucks, with 15% lower costs – delivering clear productivity benefits. They also take truck operators out of harm’s way, reducing the risks associated with working around heavy machinery.
Autonomous driving is like alphago/alpha zero: without the algorithms it's really hard to solve, but when the algorithms are ready and mature, it will be easy to do and open source.
So far the biggest winners are companies outsourcing the time of being ready to retailers: Tesla and Comma.AI
The training set, regulatory compliance hurdles and implications are a bit more complicated than teaching your computer how to play a perfect information game or even chucking text at it until it's reasonably competent at chat.
- Back of the day cab? You may as well look out of the window. Even then the following point applies (somewhat).
- Back of the sleeper? What are you looking at, front of the dry van / reefer I guess? Hardly useful that way...
- Back of the trailer? The trailer might not even be yours. I guess you could make a kit that clips onto the door handle or bumper? But then you'd be looking at... the dock gate, most of the time. Again, not very useful.
This can go on and on, but I think what you really want is a drone directly overhead.
All that is needed is a regulation that says that all new vehicles must have it. There is no technological problem in doing it that cannot be solved with today's technology.
> Back of the trailer? The trailer might not even be yours.
This is easy. Just put an Ethernet connector between the truck and the cab and put a "security camera" on the trailer. The only part of this that we don't have already is the Ethernet connector which is currently going through ISO to get a standard connector: it needs to not vibrate loose, but still break away if the trailer comes unhooked (in particular when the driver forgets to disconnect it while disconnecting the trailer)
I put security camera in quotes because you need something a little more robust that the camera you buy at best buy. This is the case, and maybe the lens, mostly it is the same thing. in any case you can buy these off the shelf today, but the truck/trailer connector won't be compatible with one from a different manufacture.
You need a lot more than just the back-of-the-trailer view. Admittedly this could prevent someone / something being run over by the rear bumper of the trailer, but that's pretty much it.
For all other cases (backup), an overhead view, real or synthesized, would be much more helpful in establishing actual situational awareness. Anything less than this would still require the driver to G.O.A.L. (Get Out And Look).
1) The model most are working toward is "hub to hub" not point to point. As in, A human driver drives the truck from the warehouse or port to a hub on the highway outside the city surface streets. Then they get out of the truck and pick up another one that needs to go back into the city. The truck then drives itself on the highway to another hub near its destination where another human driver completes the trip. This works in multiple ways - it allows human drivers to deal with the hard part of automation while also avoiding the terrible away from home lifestyle, and it allows you to have dedicated autonomous trucks for on-highway separate from your human-driven fleet (you can swap the trailers instead of using the same cab). You'll see these hubs start cropping up in a couple years.
2) On-highway autonomy is not that far away anymore. It's definitely not a "next year" thing, but mid-late decade is a near certainty based on the current state of the tech. That's a long time for people used to software, but not a long time compared to the lifecycle of a truck.
3) There seems to be an implicit assumption in this write-up that autonomous systems will remain horribly expensive, which isn't true. They're expensive today because each generation of development is effectively custom built in small quantities. Just like Starsky said their camera system would drop in price with scale, a "ready for production" generation of autonomous driving kit in a few years will likely be close to 1/3 the current kit cost, which has very large impacts on the total ROI calculation.
I don't want to make it sound like there aren't big challenges left to solve - there are - but there's also a reason so much investment money has flowed into this space. The ROI does make sense, just not on the original hype-cycle timeline that a lot of people reported.