Could it be that we don't have "dark factories" because the moment a workshop turns into an autonomous machine, we reassemble something more complex out of those machines and need again a human operator to manage it? A few decades ago, it took a print shop with several employees to print a brochure. Then came the photocopier, which does it more or less automatically as a sort of "dark printer workshop". We even let them run without supervision in a dark office room. But we still needed someone to collect the copies, bind them, mail them, etc. Then photocopier and binding machine were combined, etc., etc. More and more gets automated and happens "in the dark", but there is always a frontier were human control or intervention becomes necessary. When we cross such a particular frontier, there is typically a new frontier that opens up, while the old frontiers are hardly perceived as such any more. Think of a water pump, for example: it is a "dark factory" that replaces a human-powered well, or of the heating system of a house that runs most of the time without direkt supervision.
I'd guess that a huge of the reason is insurance, regulation and safety laws. Getting a fully automated "dark" factory would require lots of money and skill. Any company that can afford that level of automation is going to operate in a country with various fire codes and insurance requirements that probably out cost the benefit of having zero human oversight. If the humans must be around they might as well work, no reason to invest $50mil automating something the people that have to be there can do.
If there are dark factories they are in a place where the authorities don't care if it burns down at night and takes half the business/slum district with it.
I've been waiting for one of those factory simulator games to have something where when you pass some certain level in the game, it goes into Enders Game mode, and players start controlling machines in real factories. Users could even get paid to do this, and to scale up their operation with automation tools etc.
Same with farming simulators, mining simulators, fast food simulators, etc.
Neil Stephenson predicting the future yet again with reamde (sic).
The plot is a world of warcraft style MMORPG is linked to real world settings (such as airport security) to allow people playing the game to preform real life actions (in the book, the example was sound an alarm when they spotted a character travelling the wrong direction).
The idea being, if it is a game (or simulator as you say) you can make it waaayyyy more interesting than the real life work.
Between him and Daniel Suarez these "near future" sci fi writers are getting scary good at predicting some things.
Any other authors are books would be much appreciated!
Although part of the point was that this was a stupid idea, more geared to getting press coverage and investment funds than actually providing any kind of valuable way of making gameplay productive.
Daemon And FreedomTM might be one of the best near future sci-fi pairings.
I read them back in 2011ish timeframe and within a couple of years multiple things had some true from those books.
It’s ok. You’re not missing something the way you would be if you bailed on Snow Crash.
I like Stephenson but there’s always this smug vibe; I picture him writing a chapter and then high-giving himself while shouting “I so so fucking smart!”. REAMDE is high on that and lower on genuine smarts than his other works.
That may have been it. There is an early scene in some master of the internet billionaire office and I couldn't wait to put it down. Send me to the monastery on Orth.
Yeah, there are parts like Shaftoe AWOL in Scandanavia that when I reread I can scarely remember reading before, like I slept through parts of reading it.
oh no, while making sure I spelled his name right, I just got sucked into Shaftoe encountering Goto Dengo and the General on a baseball field.
There was the guy who won the NASCAR race using the trick he learned on the GameCube NASCAR game. The trick was to make the last turn skidding against the outer wall with the car floored and pass everyone.
Enders game is probably obsolete since AlphaGo zero can play better than any human. Just have the AI reinforcement learn over the course of a trillion games in the simulator.
The book version AlphaGo wouldn't have a chance at. The whole thing was a setup to get a very skilled person in charge who didn't know what was going on and would break the rules and rebel at the key time. Firing upon the planet was genocide and anyone who could give that order was inherently unsuitable for the role they must play. At the same time, it was a completely hopeless battle in which the only possible "victory" was a suicide charge against the planet (in the book the planet goes boom and take the fleet with it. It's gone, not merely a charred ruin.)
Hence one person kept in the dark and carefully prepared to be utterly ruthless at those who have attacked him (not his opponents in battle school--that's competition. He murdered (although he doesn't know they died) IIRC two bullies before the events we see in the movie) with the intent of getting him to fire on the planet.
AlphaGo would see firing upon the planet (or even at a target too close to it) as against the rules and wouldn't do it. Ender believed he had flunked out until he learned what really happened.
I've been thinking about a near future idea where people an play a real world RTS with RC cars and drones that fire small 22 rounds.
The vehicles could be assembled inside a small factory full of automated open source 3d printers/mills/pick and places/robot arms.
You could have different people playing different kinds of games with some users driving little RC tanks, or operating little turrets on stationary buildings, or driving cargo trucks that go pick up resources to feed the factory, and then others doing totally different activities like operating the factory or designing new machines for the factory to build / be built of.
I figure you could set this up in some rural state like Montana with cheap land / few gun regulations and let people pay to play, if it's successful enough your customer base will basically be doing R&D / mechanical turking your Von Neuman probe development and then when it gets small enough you can stick a couple factories on a rocket, launch 'em to the moon and make the sequel to the game be set in space...
In the US it is generally illegal to manufacture or modify any repeating firearm with a software controlled firing mechanism. The ATF considers those to be too easy to turn into machine guns. I think you would need to acquire an existing pre-ban registered machine gun, or get a license to manufacture machine guns for military and law enforcement (difficult and expensive).
Oddly enough, I’ve just recently been exposed to livestreams where people are sitting in giant racing-sim setups and controlling RC cars over the internet.
There are crane game apps with cameras pointed at real cranes so you can control a real crane live and win real prizes. I’ve never even gotten close to winning, though.
The way this looks like now (in the United States additive manufacturing industry) is that people, Zapier, and bits of scripting glue form the automation layer in most shops. This is really important, because labor costs so much more in the US compared to other manufacturing centers.
Companies that have attempted to achieve 100% automation (like YC-backed Voodoo Manufacturing) have gone out of business, because you still need people to run floor operations, do sales, handle customer accounts. What's important is to give a few generalists no-code tools, so a machine technician can quickly iterate on a model without sending it back to a CAD designer.
Here's one of my favorite examples of a 3D printing operation in the United States, if you're curious what these businesses look like. Kason Knight started iSolids with a 3D printer in a spare linen closet. He's now running a ~10 person operation out of a warehouse in Texas: https://www.youtube.com/watch?v=xPcA9uIgi7g
“The factory of the future will have only two employees, a man and a dog. The man will be there to feed the dog. The dog will be there to keep the man from touching the equipment”
Or, all the actual work done by one person like in Futurama, "My Hermes got that hellhole running so efficiently that all the physical labor is now done by a single Australian man"
Ten years ago I was at a Phillip Morris cigarette factory for a two-week IT consulting project. The entire factory was off-limits to staff during operation.
Ingredients went in one side, cartons of cigarettes came out the other.
Sometimes it seems these type of factories are so secret that we don’t hear about the, because they are… secret. Thirty years ago I heard about a “lights out” factory at Electrolux. But never really learned if it truly worked or not.
I don't think it's as much that they're secret, as that there's few people to talk about them. There's a lot fewer people interacting with them because nobody works there.
I've spent a decent amount of time visiting factories in the US and Asia, and despite "how it's made" or other videos showing a ton of automation, so much of manufacturing requires people. Until you're making millions of something or things which are super high value, automation of many assembly steps is often not economically viable. Certain steps are largely automated, like injection molding, or CNC, but so much of grabbing part A and inserting it into part B and adding screws is done by people. Toy manufacture is especially insane as so many little details are painted on by hand.
This. I have spent 30 years in working with a factory in this sort of thing--putting brains into the machines made about a 5-fold improvement in product per worker-hour (and as we added more complex products in that time the actual ratio is higher--I just don't know how much higher) and lowered the skill level required for most workers.
When they hired me it was mostly workers with hand tools. By the time my original employer was destroyed in the housing collapse it was mostly workers acting as material handlers for machines that did the actual work. I'm still in the same industry with some of the same people but now working for what used to be a minor competitor that has grown since the giant died. And note that this is entirely a build-to-order situation, not an assembly line--but it's still the actual work being done mostly by machines that know what they're supposed to build.
Working in my father’s machine tool shop taught me a similar thing about “dark” factories, lights-out shifts, and so on. For a few hundred dollars you can put a hopper after a machine and now it can process an entire bar of raw stock over 1-2 hours before a human has to tend it. For a few tens of thousands of dollars you can put a bar feeder before a machine and now it can process a half dozen bars of raw stock over 8-10 hours before a human has to tend to it. The next level is some kind of inter-machine logistics or automated packing - I would guess that would be in the hundreds of thousands of dollars range, bespoke, and call for a per-factory installation. Most importantly, this is the point where it’s cheaper to hire a human to do it.
Many factories will use bar feeders to run their machines overnight, but going further with automation doesn’t get enough return on investment.
Exactly. The human has the flexibility to do many different things even though they're inferior to the robots at any given thing. Most operations aren't at a scale where every task involved is done at sufficient volume that having a single-purpose machine to do it is worthwhile. Rather, the machines chip away at the edges, allowing the human to more and more do what they're superior at: flexibility.
It's the same thing we see with eternal claims of being able to do things without software developers--the tools chip away at the edge to a huge degree, but what they're really doing is mostly taking away the repetitive scutwork of the job, leaving us coders to do the part that takes the thinking. The building blocks I assemble these days are much bigger than when I started but the fundamental nature of the job of distilling requirements into useful directions to the computer is unchanged.
>Many factories will use bar feeders to run their machines overnight, but going further with automation doesn’t get enough return on investment.
As a rule, yes, but I say it depends on how you measure ROE and what that human costs.
In situations where there's a collective agreement in place and you have to pay that labourer $50K/year + benefits (NTTAWWT), you get more wiggle room to make your case for retooling.
Another case is when I've worked for bigger manufacturers where there were some of us on site just in case we were needed and our salaries were part of facilities overhead.
We were paid no matter what and could do what we liked as long as production was running smoothly and our minor, secondary duties were being handled. In this time (~4hr/day) we could scheme or tinker to our heart's content.
Jobber shops (a.k.a. the real world) were a different matter ...
Mazda's "Mazda 3" factory was a purpose build nearly full automation factory and that was a decade ago. I think they said it required something like only 7 people to run. I can only imagine the goal of no workers in a factory is even closer now.
Aren't all modern 12 inch semiconductor fabs fully (100%) automated? As in. The WIP is too delicate to be handled by humans, and humans are only required for machine maintenance?
Raw materials go in, and printed wafers come out.
Its usually called a lights out factory - and they some Japanese production facilities claim to run in this mode. Fanuc and Sony-Playstation comes to mind.
My personal experience with robots and factory automation makes these claims rather dubious. There is always some stressed maintainer needed, at lest on standby. Factory equipment ages constantly and even "durable" parts, like the energy chains break regularly irregularly and it takes experience to detect and pre-emptively replace these. Until that "flickering" part is replaced, you have a constant series of increasingly occurring line stops. Including, product removal and NIO product increasing. Its possible to run a busy looking factory producing nothing but scrap for days.
Broken products and its residue clog at unexpected places, a thousand parts later, the glue from the packaging it arrives in, makes the unwrapping machine sticky. Dust that comes in with the package, accumulates or static charges transport actually non floating foils to unexpected places.
Nature finds a way, and spiders build there webs in front of light or capacity sensors. Even cats bring there young into some hidden cable spaces and thats a good thing, because they prevent rats from gnawing on the cables.
Finally, the cheapest bidder wins and makes factory equipment everywhere, especially if its new, prone to breakage and failure. Resulting in the maintainers, partially rebuilding machines with self-made parts until they are sturdy. Until that stage is reached, machines can have quirks, like vibrations moving sensors of position.
Also, the cheapest supplier also comes to plc software, resulting in horrific state-machines, waiting for ghost parts that left the system aeons ago and need careful massaging by maintainers to calm the enraged machine spirits (sometimes by hitting a robot with a wrench-> It opens the safety circuit, resetting the programs base state).
Many robots are needed for very precise tasks, and need to re-calibrate in intervals to keep fulfilling there tasks. These error calibrations happen on top of the often used
Non-parametric robot calibration
Its prone to move with temperature, moisture and alot of other parameters. Requiring a adaption of the programs used in industrial automation.
Finally, there are "sales-failures", were the automation is sold, produced and then - never worked out. As in a dead "abandoned channel" of the assembly line. Its shown during factory tours as the "future" but the layer of dust and the missing traces of use give away, that it is not used in production. Usually its precision requirements that couldnt be met or would have required insane efforts. The Welding at Wendelstein comes to mind, were they created a "reference frame via laser triangulation" to prevent wrong welds, due to heat expansion of the material.
Same insanity is usually applied to car manufacturing in
germany, especially for the
https://de.wikipedia.org/wiki/Spaltma%C3%9F obsessions, which regularly result in insane bake offs, with all kinds of robots, sensors and suppliers. That sort of machine just throws alarms, just from heavy equipment working nearby. No lights out there.
Ever.
Some guy sitting nearby, hitting the acknowledge and retry button after viewing up from the cellphone. So anal retentive quality control is a direct opponent of lights out, they want lights on all the time and fast detection of creeping in errors.
So, its a nice goal, but until machines can handle all of the above. No.
The only factories, that by "need" and not additional requirement are "lights out" that i have ever experienced, are chip factories, in particular the etching, photo-masking, implanting, etc.
Its only when non-automation become a "destructive" force, by lowering the yield, with extreme costs, lights out comes to fruition. And even then. The additional costs are staggering. All technology is not "off" the shelf industrial equipment, but special designed equipment to reduce wear, tear and particles.
Another case is, were the equipment or process is dangerous to humans. Battery manufacturing or chemical industry, come to mind. Usually the lights out there is a Encapsulating layer around the machinery, with close direct observation by human overseers.
Yup. I've seen a planned computer assistance step get abandoned because we never figured out a way to work around thermal expansion issues. The objective was to laser-paint a drawing of what they were to build, but we couldn't keep the laser from drifting. (That was ~20 years ago--now I would solve it by putting a camera with the laser, noting the discrepancy and adjusting the laser.)
Thank you for providing this grounded view. Writing software, the moving parts seem to work almost perfectly except the rare bit-flip. But it seems factory-scale robotics are not like that..
I was gonna say, at any given facility you'd need at least a man and a dog: the man to feed the dog, and the dog to make sure the man doesn't touch anything.
> “Right now, we’re more in a state where companies are less interested in implementing the full dark factory concept and more focused on how automation can complement labor, and this nuance gets lost in the excitement of robots,” he said.
Right now, nobody wants to stand on the top of the mountain.
It looks like everyone wants to scale the mountain as fast as they can. But in fact, their primary concern is really just how to take their next mundane step!
> Even if a fully automated facility exists, you probably still need at least some full-time employees to maintain that automation. So, is that truly a dark factory? Under the definition, no.
If any maintenence by a human means it's not a "dark factory", then they definitely don't exist. That seems like a bizzare definition however. People aren't claiming they have self maintaining facilities.
I worked for Kodak the summer of 85 as an intern. I've shared interesting things about that on HN before. But this just reminded me of something else. The Elmgrove facility where I worked had 15,000 employees and dozens of large buildings.One of the very large buildings had no windows at all. When I asked about it, my manger said that it's all robots in that building.
How I've wanted a container shop which builds what I have with stuff that is received by the container robots & assembled within for shipping it out in another country completely!
One might note here that not all robots are created equal. Robots dependent on machine vision to function would need a lighted factory similarly to humans.
You also can’t let the temperature drop too low or the lubrication for all the gears and servos will solidify. Even robots have running temperature requirements. In colder areas, you have heated storage because lots of electronics don’t like being stored in freezing conditions.
CNC adoption is actually driving improvements to the physical plants I work with--the machines can't run with it as hot as people can so they're starting to actually put in AC. Oh that and the old guys that work in those machine shops can't deal with the high temps like they used to so management is afraid they're going to pass out and fall into the machines...
...but not necessarily the same climate. Depending on the robot, they could have a higher MBTF if the thermostat were set at 0C, or 50C.
And more interestingly, there might be value in isolated sections of the factory being climate-controlled to very different temperatures, depending on the particular robot in use. Not like these things walk around between stations.
Most 'visual servoing' systems need edges or contours but not colour.
In the long run it would still probably be cheaper to have infrared cameras or a few LED rings around the camera lense rather than full on industrial grade lighting 365 days in the year.
I wish they would skip the factories and go straight to the fast food industry. I am tired of low paid teenagers ruining my fast food order!
I look forward to our robot overlords!
You take this concept far enough and factories become like corporate bonds. Put money in, money comes out. The question ultimately becomes: if no-one’s labour is needed, do we need to retain money?
>if no-one’s labour is needed, do we need to retain money?
If everything that people do to maintain society is called something other than "labour", then I guess it would equally make sense to call the accounting concept that tracks it something other than "money", but what's the difference?
Not all labour is performed in factories, so... yes? It's hardly be feasible to have "dark hospital" for example. Even if the robotic doctors would be fine without light, the patients would be a lot happier if they can see.
