16 votes

Where are all the robot trucks?

Topic removed by site admin

16 comments

  1. [9]
    DeaconBlue
    Link
    I am not excited for the next cycle of "new technology is scary" in regards to fully autonomous trucks. With electric cars, every single time one of them caught on fire we got headlines about how...

    I am not excited for the next cycle of "new technology is scary" in regards to fully autonomous trucks.

    With electric cars, every single time one of them caught on fire we got headlines about how scary that concept is. "How can we get into cars that can randomly burst into flame?" people ask, while ignoring that any given day driving through a major city you could see two or three cars that burned down on the side of the road.

    Then we got the autonomous vehicles (that might have gotten banned again? I am struggling to keep track) in California, where every accident involved with them got headlines of how awful it was that a bad thing happened when machinery interacted with humans.

    And next up is going to be the worst wave, trucks. Any accident with a truck is going to be gruesome. There will be some. There will be headlines asking how we can accept this risk to society. People are going to get run over, drive their cars under trailers accidentally, or otherwise mangled. I know this because it happens every single day and nobody cares.

    All of the hype and the headlines will make it difficult if not impossible to actually compare the efficacy of the autonomous vehicle compared to Joe Truck Driver until much, much later.

    29 votes
    1. [5]
      vektor
      Link Parent
      The fun part is that many of the failure modes of automatic cars are crassly different from human-controlled cars. Driver asleep? Man that is scary as all hell, but it happens all the time. People...

      The fun part is that many of the failure modes of automatic cars are crassly different from human-controlled cars. Driver asleep? Man that is scary as all hell, but it happens all the time. People don't care. Computer vision interprets a harmless thing as a threat and emergency brakes? Stupid computers can't even get the absolute basics right. It's too fucking easy to portray them as dangerous, if you're into that, irrespective of the facts. But the facts are what we should use to orient ourselves. Does manufacturer X's car have a better or worse accident rate than humans? It's a straightforward question, but the methodology doesn't follow in a straightforward way, as there's many kinds of accidents with unclear weights as to their gravity. Nevermind that the actual fact-finding that then needs to happen is often difficult.

      It's also not helpful that there are drastic differences between different companies. As far as I know, Tesla is downright reckless with the self-driving practices for reasons that I think we've all heard; compare that to Waymo who have been pioneering this tech for more than a decade now and have basically committed to running their fleet only in well-mapped "walled gardens", which is an intrinsically safer strategy. It's dishonest to judge Waymo by the effects of Tesla's actions.

      And a different safety environment awaits us again with trucking, particularly long-distance. Hell, you can restrict these trucks to highways and truck stops and just have human drivers drive them the last miles to the destination. This way, the trucks only really need to be able to navigate a very simple environment. Much easier to solve. To expect equivalent safety results from all three approaches is a bit silly.

      I've opined before on why I think self-driving personal vehicles will lead to less road use overall, as a ridesharing or renting model becomes more possible, relieving drivers from the economic pressures of ownership. I can't say I see the same effect for trucks, but the efficiency gains are undeniable; it's just a matter of if they're fairly distributed. Furthermore, there are secondary effects: You could for example coordinate several self-driving trucks to drive in a very tight convoy, reducing air resistance substantially, saving a lot of fuel, and freeing up road space. And sure, that's an imperfect approximation of a train, but at least it's more efficient than existing trucks.

      13 votes
      1. [4]
        scroll_lock
        (edited )
        Link Parent
        Someone made a joke to me recently about every transportation "innovation" eventually looking more and more like a train, apparently like evolutionary biology and crabs. I can't comment on the...

        And sure, that's an imperfect approximation of a train, but at least it's more efficient than existing trucks.

        Someone made a joke to me recently about every transportation "innovation" eventually looking more and more like a train, apparently like evolutionary biology and crabs. I can't comment on the crabs, but the joke really is true... there is no system more energy-efficient at transporting goods over a set distance than trains. There probably can't be, because no amount of automation can overcome the fact that rubber on asphalt is fundamentally far less efficient than steel on steel; and that train capacity literally cannot be beaten by anything else. While freight trains typically run a bit slower than trucks, technically they can run at 65+ mph as long as investment is made to the infrastructure and rolling stock (see: Super C and others). (Technically, a small, lightweight, electric freight train could run just fine at 220+ mph, it just needs a track.)

        There is no reason that trucks ought to be so prevalent in long-distance shipping in the United States. That they are permitted to do so is a massive policy failure resulting in higher gaseous emissions, higher tire particulate emissions (toxic), extremely high infrastructure costs (road damage is caused by heavy vehicles, and the heavier the vehicle, the more damage... and it's way worse than linear), and contributions to traffic (no matter how much you try to automate it). An automated truck might be safer than a human-operated one on a highway, but not necessarily much in a city or on an arterial, so I'm not going to count on that being a particular benefit; having human drivers just for that last section doesn't solve the safety issue. Railways can provide long-distance shipping in a way that has far fewer social and environmental externalities: rail tracks have a much smaller physical footprint, require fewer materials, and do not degrade at the same rate that highways do; trains likewise always emit fewer particulates. In fact there is little reason that large trucks are necessary for short-distance shipping either—this is also a policy failure, and such vehicles ought to be replaced by smaller delivery vehicles, especially in urban areas, where golf cart-sized electric bikes will do the trick. Truck shipping is economically competitive because of its negative externalities and because of the ridiculous amount the government invests in highways instead of freight rail systems.

        why I think self-driving personal vehicles will lead to less road use overall, as a ridesharing or renting model becomes more possible, relieving drivers from the economic pressures of ownership

        I can see why this might seem likely, though I am skeptical. Automation in this sense might decrease car ownership rates, but probably not usage. Endlessly orbiting automobiles contribute significantly to traffic (current rideshare systems carry nothing and no passengers 40% of the time), and their availability itself induces "demand" for car trips that otherwise would be made, usually, by public transit, cycling, or walking (not just driving). That causes more traffic and, virtually unavoidably, more pedestrian deaths. The only way to reduce this wasted orbit time is to run ride-hailing vehicles on fixed routes, which defeats the purpose. Or to "bundle" riders with each other... which they've tried, and which doesn't really work on non-fixed routes. Taxis are inherently inefficient, a traffic engineer's eternal foe.

        In an urban or suburban area where automated taxis would be economically competitive, public transit is probably always going to be the more economical option for an individual, even if bad infrastructure prevents it from being the fastest. I made the regrettable choice to take a Lyft this weekend a distance of about 15 miles which cost $40; an equivalent train ride would have cost $5. This was an off-peak time. In hindsight, I would have rather been at my destination 20 minutes late and saved the $35. (I took the train back.)

        I don't think that companies are going to be sending their automated taxis around to rural or exurban areas, where people are currently the most car-dependent, because despite high car usage in those areas, there are comparatively few potential customers. At least they would not do so at reasonable prices. I think that it will be possible to rent automated cars in the future but I don't see how this could reduce driving overall. It's still a car. The way to reduce Vehicle Miles Traveled (VMT) by car is to give people options that aren't cars: pedestrian/transit/cycling infrastructure.

        6 votes
        1. [2]
          Comment deleted by author
          Link Parent
          1. scroll_lock
            Link Parent
            A lot of what would probably otherwise be rail freight transport in Europe is done by maritime vessels. This is less common in the United States, which is less coastal. In the US, ground...

            A lot of what would probably otherwise be rail freight transport in Europe is done by maritime vessels. This is less common in the United States, which is less coastal. In the US, ground transportation dominates. Historically the biggest US cities were all by major waterways navigable by gigantic ships, but this is less true as more people move to inland places like Houston, Dallas, Phoenix, Las Vegas, and Atlanta.

            If you want to get something from LA to NY by sea, it has to go through the Panama Canal, so rail or truck might make more sense. But going from Lisbon to Istanbul is comparatively straightforward. It's not like Europe doesn't have reasonable freight railways, but many more of their population centers are coastal, so there's more competition against ground transportation as a whole.

            I agree that more can and should be done to improve rail infrastructure in both regions. There are various intergovernmental projects in the works in Europe that will improve freight access across the continent, notably some rail tunnels through the Alps to ease congestion. Lots of stakeholders in the EU have the specific goal of reducing reliance on trucking as a mode, mostly for environmental reasons. That goal is basically nonexistent in the US, though it isn't really any less important here.

            4 votes
        2. [2]
          vektor
          Link Parent
          Flexibility. Whic makes me curious about more flexible train systems. Maybe one could use the most bare bones self driving tech to make rail cars that can automatically rearrange into trains. That...

          There is no reason that trucks ought to be so prevalent in long-distance shipping in the United States.

          Flexibility. Whic makes me curious about more flexible train systems. Maybe one could use the most bare bones self driving tech to make rail cars that can automatically rearrange into trains. That feels like the big bottleneck to getting stuff across the country quickly. But it in a shipping container, truck it to a rail yard, now you gotta wait for the next train to wherever. Realistically any train going in vaguely the right direction would be fine, but detaching and reattaching rolling stock is so slow, dangerous, manual and ultimately expensive that you don't do it, so now you're stuck unable to reasonably send individual containers in a timely manner. As a result, only companies that can fill a substantial fraction of the entire train will ever use it.

          If instead the train car knew where to deliver its container... it could join a 'convoy' of other cars wherever convenient, and maybe even go to the final destination entirely on it's own.

          1 vote
          1. scroll_lock
            Link Parent
            Trans-shippment is indeed the economic reason companies choose the much more environmentally harmful and ultimately socially expensive method of truck shipment over rail in many cases: because the...

            Trans-shippment is indeed the economic reason companies choose the much more environmentally harmful and ultimately socially expensive method of truck shipment over rail in many cases: because the cost to the company is lower when they can shift it to the public.

            Any automated technology we see on roadways can probably be more easily replicated on train tracks, which have fewer external conflicts and therefore need far less complex detection mechanisms. Some self-driving freight trains already operate successfully. This technology is fairly common in modern passenger trains, mostly metro services, including in the United States and Canada. It's generally more operationally efficient even disregarding the financial cost of employing people. I would expect to see more such trains in the future.

            The way classification yards are organized is pretty complicated and to some extent relies on gravity, but I don't see any reason why it would be incompatible with automation. As you speculate, if a traincar "knows where it's going," it could reconfigure itself more effectively in a yard. That's a headscratcher for an engineer for all the reasons doing it manually is, except that now humans aren't supposed to be involved. I don't have the expertise to comment on automated sorted mechanisms in much depth. The yards and the traincars themselves would need to be significantly upgraded to communicate this way.

            You don't really need to get the train to its absolute final destination, just to an area depot or factory where last-mile transport (probably road transport) can pick up the slack in an intermodal loading yard. That last-mile transport can be a lot of things and certainly does not have to be a bunch of 18-wheelers. No single company strictly needs to be filling orders for an entire long-distance train, though they can.

            It's worth noting that rail workers in the United States are unionized at a higher rate than truck drivers, so there might be more political resistance from automating away their jobs. But to be honest I think it would be better if freight companies allocated fewer resources toward operating trains and yards and more resources toward rail inspection, which is something those workers could easily pivot to.

            3 votes
    2. first-must-burn
      Link Parent
      Having worked in this space (autonomy safety) for about ten years, if you want straight talk (not corporate spin) about safety in the autonomy space, I recommend following Phil Kooopman and Missy...

      Having worked in this space (autonomy safety) for about ten years, if you want straight talk (not corporate spin) about safety in the autonomy space, I recommend following Phil Kooopman and Missy Cummings.

      12 votes
    3. devilized
      Link Parent
      Classic availability bias. And it is indeed really difficult to find information about the efficacy and safety of autonomous vehicles. Tesla published their report, which obvious will be spun in...

      Classic availability bias. And it is indeed really difficult to find information about the efficacy and safety of autonomous vehicles. Tesla published their report, which obvious will be spun in their favor. And if you google for statistics on autonomous vehicle accident rates compared to human, the first result that comes up for me is a law firm claiming that autonomous vehicle accident rates are 2x that of human drivers.

      Each of those organizations have a strong bias in their own direction, and therefore has a lens that they want to view this data under, and will spin the numbers to tell the stories that they want to tell. It's going to be very difficult to obtain this kind of information.

      5 votes
    4. [2]
      Comment removed by site admin
      Link Parent
      1. steezyaspie
        Link Parent
        Yeah, I can count on one hand the number of times I've seen a burning car on the side of the road, I don't think I've ever seen a burned out husk just sitting on the side of the road.

        Yeah, I can count on one hand the number of times I've seen a burning car on the side of the road, I don't think I've ever seen a burned out husk just sitting on the side of the road.

        1 vote
  2. [3]
    Idalium
    (edited )
    Link
    You are the CEO of DeliveryCorp. You have replaced all of your drivers with machines. This is good; everyone knows that they are safer, and fewer accidents means less paperwork. Every time your...

    You are the CEO of DeliveryCorp. You have replaced all of your drivers with machines. This is good; everyone knows that they are safer, and fewer accidents means less paperwork. Every time your automated vehicles make a delivery, the company makes money. This is all very good.

    But you didn't become CEO by being satisfied with 'good'. If you made more deliveries, you'd make even more profit. So why not tweak the AI a little? Make them drive just a little faster, and take just a few more risks. You could make 2% more deliveries per day.

    So what if there are also 2% more fatalities? Nobody's really accountable, and it's still better than when you had an overworked, sleep-deprived human workforce. You can just pay the compensation to the victims; you're making more than enough extra money for it.

    If it works, why not push it further? 3%, 4%, 5% more deliveries. You can keep going! Eventually, you'll reach a sweet spot where your fatality rate is ever so slightly better than when you had weak, fallible human drivers (according to the studies you funded).

    Sure, if anyone actually notices what you're doing, there might be a brief social media storm, but the kind of people who care weren't your biggest customers anyway. You'll blame factors outside your control, like population density, or bad parenting. At worst there might be an inquest, you'll throw a software engineer under a bus (so to speak), and the company will have to pay a fine. You've already ringfenced the fund for that.

    After a decade or so, people will forget that human drivers (and the concept of individual accountablity) ever existed. Then you can push your AI to optimise profit as far as you like!

    After all, it's not like you'll ever go to jail.

    5 votes
    1. Idalium
      Link Parent
      That post is a fun piece of creative writing, not a carefully-researched manifesto.

      That post is a fun piece of creative writing, not a carefully-researched manifesto.

      8 votes
    2. EgoEimi
      Link Parent
      Good news! A new position has opened up at the company. :)

      At worst there might be an inquest, you'll throw a software engineer under a bus (so to speak), and the company will have to pay a fine.

      Good news! A new position has opened up at the company. :)

      2 votes
  3. [4]
    boxer_dogs_dance
    Link
    I see three issues that weren't addressed by the article. They are all potentially solveable, but these are my questions. One is weather. Robot driving in the desert southwest is different from...

    I see three issues that weren't addressed by the article. They are all potentially solveable, but these are my questions.

    One is weather. Robot driving in the desert southwest is different from driving where there is ice, blizzards, heavy fog, blinding rain storms.

    The second is security. I am convinced that if these unmanned vehicles carry anything of value that organized and disorganized crime will try to jack the trucks, either through hacking or using low tech methods, blocking the vehicle's path and taking the cargo.

    The third is maintanence and repair and communication when something goes wrong. If a tire blows in the middle of nowhere, who communicates with local authorities? Who replaces the tire? Who gets the truck off the road?

    2 votes
    1. turmacar
      Link Parent
      I generally agree. But at the same time the first gen of something doesn't need to be a universal solution. If robo-trucks can't handle snow and adverse weather, they'll be adopted down south...

      I generally agree. But at the same time the first gen of something doesn't need to be a universal solution.

      If robo-trucks can't handle snow and adverse weather, they'll be adopted down south where/when it's viable and not up north where it's not. Maybe if there are less routes in more demand truckers will be in a better bargaining position for better pay/etc. Though that seems unrealistic without a union of some kind.

      For security/maintenance/repair, that's more of an argument for trucks returning to being the 'last mile' delivery solution instead of long-haul, and rebuilding infrastructure and modernizing railways. But we in the US seem chronically opposed to that.

      5 votes
    2. [2]
      vektor
      (edited )
      Link Parent
      As for security, I can actually see an argument for why autonomous trucks are better: They are chock full of sensors. All it takes is a few lines of code to make sure the footage gets backed up...

      As for security, I can actually see an argument for why autonomous trucks are better: They are chock full of sensors. All it takes is a few lines of code to make sure the footage gets backed up when the truck observes certain anomalous conditions, i.e. being forced to come to a full stop by a car directly, and you have all the surveillance footage of the event that you could want. Dispatch then looks at it, calls the police and perhaps calls upon other AVs to track the getaway vehicle.

      I don't think a truck driver will pull out their coach gun and defend the goods. They'll get the tags of the vehicle involved and call the police, and that's it.

      4 votes
      1. boxer_dogs_dance
        Link Parent
        You might be right, but law enforcement is already overextended most places. Ski masks and fake license plates are not difficult to acquire. It will be interesting to see what happens.

        You might be right, but law enforcement is already overextended most places. Ski masks and fake license plates are not difficult to acquire.

        It will be interesting to see what happens.

        1 vote