If we want to grow US economy by making existing humans more productive per capita, then we need cheaper energy per capita, cheaper intelligent robots per capita. Also goes to say we need less NIMBY-ism per capita.
We need high speed rail building robots, solar farm building robots, home building robots, healthcare robots. Cheap repairable open robots. Robots that defend us, robots that transport.
More robots than humans. More robot companies diverse in their approaches.
What will humans do in that setup? I always fail to understand how replacing labor by capital in our current system helps for prosperity once a certain threshold of productivity is reached.
Sure selling your physical labor isn't ideal, but it's better than having no wage at all because there's nothing in your reach that isn't done by robots. We won't have a basic income, Americans can't even get a proper healthcare...
I guess theoretically humans do more complex work? Rather than screwing in a solar panel, humans supervise the robots that screw in the solar panels. Then in theory we'll have the same number of humans employed in the solar industry but 10x the solar installed per year. When spreadsheets came along we didn't have less accounting work, we had infinitely more. That being said I am very skeptical that AI will be the same, or that our society is in any way prepared for a rapid transition away from human labor.
Humans working on farms with tractors are more productive than humans working with oxen to till land.
Accountants working with computers are more productive than accountants working with books and a hand-held calculator.
There will be a Pareto curve of distribution.
If we want to reduce the cost of things, we have to reduce the cost of inputs and make the markets transparent (while holding a base level of quality and safety).
This is why TVs and electronics become cheap while healthcare costs balloon out.
There's a country in Europe currently being invaded in a war of conquest because the invader can for the explicit purpose of subjugating the population and seizing territory and natural resources.
Maybe from attacking others? So each time someone calls for a war or some such action they would be shot automatically by AI controlled robot with gun. Seems effective and reasonable solution.
Movies are fiction. You can't make policies based on fiction. You could make policies based on real concern shown by the movies, but they have to be based on reality.
I always go back to GM and its first welding robots. They had to make changes to the construction and engineering process to optimize for those welders (and for them it was slow).
I assume that the current install process is fairly "human optimal". How much of that changes with your product to make it "automation" optimal? Were there small changes here for big wins? Or are you still using the same process with complex automation (the GM way), and do you see challenges arising from that?
One of the coolest parts of our job is figuring out what tweaks we can make to the process to make things more robot-friendly!
One thing we're already doing that's different from human installers is the bay pre-assembly approach you can see in our demo video, where we build chunks of 5-9 modules on a tube and carry them out to the field. Ordinarily humans put up all of the tubes, then all of the brackets, then all of the modules. The pre-assembly approach allows us to do a bunch of work in a more controlled factory environment on a different part of the site.
Longer term there are other ideas we have, but since everything has to last for 30-40 years out in the field you have to be very strategic about making the big changes...
There was User Experience (UX) and then Developer Experience (DX). Now there could be Robot Experience (RX). I wonder which of the same principles would apply.
I think of this as Rosie Jetson robotic development. In the Jetsons cartoon show, you can watch the robotic maid, Rosie, manually washing dishes like a human. Today, we have a dedicated device purpose built for dish cleaning that operates nothing like a human. What is the end state of a machine which can operate more efficiently without being built around the human pattern?
(Although, Wikipedia is telling me dishwashers already existed at the time of the Jetsons, so my analogy is bad, and I should feel badly).
If I want to wash 100 dishes I'll take a dishwasher. But if I want a single dish washed followed by watering a single plant followed by taking out the trash I'll take Rosie.
Humanoid robot has much more Jack of all trades potential.
There is no generally-agreed definition, but I would use the word "robot" only for machines that have manipulators (i.e. some equivalent of arms with hands; something similar to octopus tentacles or elephant trunks also qualifies as a manipulator).
Originally "robot" was used only for humanoids, but the only essential capability that distinguishes robots from earlier machines are the manipulators, so in all other respects they may not resemble humans or animals nowadays.
Maybe “an automated thing that moves to do stuff”. Generally there are manipulators and mobile robots like vacuums. An interesting edge case would be a CNC machine which has degrees of freedom similar to a pick and place robot but is seen as a whole static thing with moving parts. Also if the mobile platform transports people it’s usually not called robot.
This looks pretty awesome, but any reason you don’t use these in a central location to make foldable preassembled racks, instead of transporting the factory to the site?
Seems like you could fold them like an accordion into whatever max size can be carried by a forklift, drop them in place, pull them out to unfold, screw the whole prewired assembly down with ground screws, and plug it into the neighboring assembly?
This would probably work best with E/W racking, would go from tightly packed to corrugated.
I’m sure there are challenges I’m not considering, but with modules getting insanely cheap, it seems like racking/assembly/inverters are starting to dominate, and scaling up from single modules to full assemblies assembled in factories seems like a good next step.
(I’ve been building my own 20kw ground mount array, so I’ve had plenty of time to daydream about something we could’ve just dropped and pulled out)
Hey thanks! The trouble with the origami racking approach is that the engineering decisions you make to build a portable, foldable racking system go in some different directions from what makes the best hardware for long-term, large-scale installations.
A couple of ways this shows up: 1. making your system rugged to handle extreme weather. These sites need to last 30-40 years without breaking, which pushes you to use pretty heavy + large components. Piles get driven several feet into the ground, so even with unfolding systems you're still probably pile driving. 2. tracking -- adding the complexity to enable rows to rotate to face the sun makes it harder to make the system portable. Basically all large-scale solar getting built is single-axis tracked. 3. module cleaning -- it's important that vehicles can traverse rows to clean dirt off of modules on a regular basis. Having structure that connects the rows together may make this more difficult.
None of these are intractable on their own but together they make the approach you describe very difficult.
Thanks for going through that! We used aerocompact, with 18” ground screws/anchors (a lot of them), and their engineering plans for our anticipated wind load didn’t even require all possible ground screw spots to be filled, so it seems like piles might not be strictly necessary?
With panels getting so much cheaper, is tracking necessary still? Maybe if land is at a premium, but I know the aerocompact commercial deployments don’t support tracking.
Yeah, we used the g20 south-facing ground mount, no ballast, about 40 18” ground screws for a 48 panel array (relatively small 108 cell panels). Soil is clay, for reference, not very rocky, so might be unusually well suited for the screws. Easy to drive with a socket wrench or impact wrench, 1” heads.
30+ years sounds excessive for managed installations (anything thats not just sitting inaccessibly on a roof with prohibitively expensive access costs), would think max 10-20 years, if you have the option to upgrade some higher efficiency panels for maintaining best ROI given current trends continue and land being a cost factor in most regions.
Your description almost exactly matches the Maverick solar array produced by the Aussie company 5b (https://5b.co/) - you are definitely on to something.
Factory assembly and field deployment is a far faster, safer, and higher quality method of building solar farms.
Ha awesome, that looks almost exactly like what I was imagining. I guess it was somewhat obvious to anyone who’s used racking like the one I used and thought about it. I wonder how well their system works in practice, but I can’t imagine it’s less efficient than positioning/attaching these things and wiring them up in the field, and then crawling under and tucking tons of wires away with painful little clips.
Now that that’s taken care of, we just need some way to get the interconnection queues moving faster :-)
> In some cases, solar construction companies were turning down projects because they couldn’t find workers. In other cases, workers were commuting for hours to get to sites with no hotels or grocery stores nearby.
So ... tired ... of ... this.
This is NOT a shortage. This is lack of salary. The oil industry somehow manages just fine to get workers to really out of the way places.
Apparently the companies aren't drowning in enough projects to pay more money.
Modules for solar farms have been getting larger and heavier, but the maximum size/weight has traditionally been limited by what human workers can heft into place. I'm interested in what the optimum size ends up being when those human limits are removed.
What's the most labor intensive part that remains when robots are putting the modules into place on racks? Attaching cables? I would guess that robots still aren't dexterous enough for that.
I doubt the size limit would be removed. Even if the system is installed by robots, it's presumably going to be maintained by humans. If a panel dies early or gets damaged, you want to be able to swap it out. There are also economy of scale benefits for maintaining a common architecture with other solar applications, like rooftop installations.
Rail systems could and if you look for it, already do form gutters along the panel-to-panel seams to not drip water inside the contiguous array area.
Still cheaper than a metal roof intermediate layer, though.
But in places like Germany overhead glass outside of some limited applications (notably greenhouses for farming) isn't allowed to drop on anyone's head just from a sharp pebble impacting and cracking a glass layer. Sadly the regulations are just as strict for a home carport/garage-without-door as for a majestic train station hall. So aside from the worse cooling it's in practice cheaper to just use a corrugated sheet metal layer to catch rain and falling glass, with the structural support for both unified.
For us, we're happy for modules to get quite a bit bigger since the time it takes to build a bay (what we call one assembly of a few modules on a metal tube) is approximately proportional to the number of modules, so bigger modules => fewer modules => faster build times. Our robot arms could handle much larger modules.
For traditional human installation, we're already approaching the limit of reasonable module sizes -- today's modules take two people to lift and align.
> Modules for solar farms have been getting larger and heavier, but the maximum size/weight has traditionally been limited by what human workers can heft into place.
Are there installers that actually use humans to lift solar panels onto racking on solar field projects? I figured everyone was using vacuum pump panel lifters instead of humans these days as it’s much faster and less prone to injury.
I'm trying to find recent Time Lapse videos - here is one from Australia, 2020 - at the time the largest in Western Australia - so at scale. All the panels were hand lifted by one person. https://www.youtube.com/watch?v=uCtFG-hIHdM
- 200 MW
- 1200 Acres
- Piles -- 68,350
- Support Piles (Crew does 80-100 piles/day)
- Tracker System
- 490,150 Solar Modules
- 540 watts/module
- Crew can install 500 panels a day
- These are multi person 2-4 people lifting.
- 500 workers on site - doing 6200 panels/day.
- (That suggests ~40 person crews if all worker installing
panels. Wow - now I see why this is a big deal)
- Cables panels -> Inverters
- Big Inverter install to to convert to AC
- Lots of concrete pouring for substation/Inverter Pad.
I would imagine that the next constraint in the size would come from the supply chain, so you'd probably end up seeing panels that are 4'x8' if human liftable weight wasn't a consideration.
Perhaps the next size up would be the interior dimensions of a cargo container so they could slide in/out like sticks of gum.
Every time I see news of a robotics company reportedly disrupting an industry, all I can see is the very large amount of prep work using conventional techniques (read: humans in excavators) to prepare the terrain for the (usually very strict) constraints of the robot being publicized.
If the new robot can install panels very quickly, but you need 3 weeks to flatten the field to unreasonable tolerances or install tons of dedicated framing and supports for it to function, your efficiency gains or cost saving numbers will take a hit.
Those considerations are unsexy and boring, they put a shade to the press release, yet are very important to assess the value of the innovation. I'll trust a company when they start being transparent about these aspects.
These folks built a robotic restaurant. It took the form of a food truck in a shipping container, and they put it in our little food truck area here. I then saw for the next 3 months, engineers frantically running around a semi operational thing, until they took it away because it didn't work.
The investors could have just invested in some food trucks, and made money.
There already exists tight constraints on field grade. The mounting hardware has limits to how much one section can differ from the next. You can imagine that you need each horizontal beam some maximum number of degrees from parallel, and the vertical posts will only have so much extra length to account for variation.
i'm down with solar robots, but this is definitely one of those problems that's only a problem because of overly restrictive immigration laws. installing panels is pretty easy, there are tons of people who would be delighted to come to the US and install for $20/hr.
source : have a solar engineering firm in Kenya, and have never had problems with panel installation labor (QA is another story).
Try talking to a landscaper anywhere in the desert southwest and see for yourself. Not the guy who has his name on the truck because there are a few natives who still own the business because inglesa, talk to the guys actually doing the work out in the sun. 100% Hispanic immigrants.
On the domestic side we need go make rooftop solar more efficient to install for countries like the UK, there aren't enough people to do it there either.
Broadly, I think there's so much potential for computing in construction. If we can model logistics decisions as optimization decisions, we can leverage incredibly efficient solvers to extract efficiencies. The challenge, of course, being the "info pipeline" - integrating machines with construction processes and planning software and orchestrating it all in a way that adds value.
If an entire construction plan is modeled, you could also compile it into a staged procurement/shipment plan, and essentially orchestrate the supply chain in sync with the jobsite activities.
It'd be cool to have more info on how the process looks like with/without these machines, how rates of specific activities are impacted etc. but a bunch of this stuff is probably proprietary.
The challenge is that nothing gets built strictly according to the plan. Solar farms seem easy because there's no foundation, everything is above ground, and there's essentially no finishing involved.
This really is the "lowest hanging fruit" of the construction industry.
There are foundational elements, those arrays of panels do have to be mounted to the earth somehow after all. I've built solar farms where we drilled thousands of 5-foot deep, 6-inch diameter holes with 1/4" tolerance for the location. Into which pipes were concreted and the long crossbars for the panel arrays were installed.
Did the tolerances truly need to be so tight? Probably not, and just set so by a designer somewhere. But the panels do have to line up for tying into conduit that goes to (pad mounted) transformers and distribution equipment, etc.
Sure, but in that case, the alignment really just saves you on conductors, it's not required for the larger structure to operate correctly. The posts aren't meaningfully tied together in any structural way unlike a post and pier "foundation."
What I meant was, for something like a building, the plan will describe the invariants of the design, but what it really reveals are all the available variants in achieving that outcome on the actual site. With a solar farm that available variance is actually very small because the overall "plant" has very little additional or emergent structure.
This is somewhat like AVX - it only works if you have large enough vectors (a simple operation that is repeated a million times on contiguous data). Arbitrary buildings are much harder to optimize.
For comparison, here's a nice video (in Brazilian Portuguese, but should be understandable even if you don't know the language) showing a solar farm being built in the traditional way: https://www.youtube.com/watch?v=_W1nQT7az8c
Would it ever make sense to pour foundations of some kind rather than put in poles? I'm thinking maybe a trench making machine could be used and concrete poured into the trench, and then the solar panels could be secured by putting a wide base into the concrete instead of a pole?
My good buddy, Union carpenter, laughs at Silicon Valley hipsters trying to replace his industry with robots.
One of his greatest quotes is ‘We already have 3d printers for buildings, they’re called cranes’.
His advice every time I send him one these articles usually rounds to: ‘If these smart@$$ %|~>heads would spend a couple of weeks on the job actually sweating and doing some real labor, they’d realize the problem isn’t lifting things, it’s organizing it.’
I mean he can laugh but then one day it'll have happened.
For one thing, he's already identified but failed to realise the problem: organising labour to be on site and ready to work is expensive and tricky.
But when there's less labour and more machines, you're now only paying opportunity cost, not direct costs of people turning up and being unable to complete the job because it's not ready to go yet.
And unlike people, the quantity of capable machinery can constantly increase and thus the cost per unit decrease.
Eventually the machine is cheap enough to leave on site even if something else delays it.
What's the turn on capital for a buyer? Like, how many panels does a construction company need to install to pay for your system, and how does that line up to project size?
This seems like a great construction robotics application since you have such an empty / controllable site. Did you ever consider teleoperations as well?
Seems like it assembles individual panels into bigger blocks... But why do that on site? Surely you can do that much much easier in a factory somewhere?
What's the name of the solar company that just lays down solar panels flat on the ground? Ersolar? That's the future of solar Farms. panels are just cheap enough to cover the ground flatly without complicated buildings.
This thing mounts panels and screws them in. So? That is maybe 1% of the needed labor, and could be more easily optimized through better connectors (snap-on or drop-in panels). Show me the robot that can survey and install foundation posts, wire up the panels, or clear brush and open the pallets of equipment during initial setup. The robot revolution is coming, but this isn't it.
We need high speed rail building robots, solar farm building robots, home building robots, healthcare robots. Cheap repairable open robots. Robots that defend us, robots that transport.
More robots than humans. More robot companies diverse in their approaches.
Sure selling your physical labor isn't ideal, but it's better than having no wage at all because there's nothing in your reach that isn't done by robots. We won't have a basic income, Americans can't even get a proper healthcare...
Accountants working with computers are more productive than accountants working with books and a hand-held calculator.
There will be a Pareto curve of distribution.
If we want to reduce the cost of things, we have to reduce the cost of inputs and make the markets transparent (while holding a base level of quality and safety).
This is why TVs and electronics become cheap while healthcare costs balloon out.
Defend from who, and who is us ?
There's a country in Europe currently being invaded in a war of conquest because the invader can for the explicit purpose of subjugating the population and seizing territory and natural resources.
We support that Terror state too, Israel.
Do you not know of all the technology that we enjoy today that had movies and books decrying the dangers?
https://www.youtube.com/watch?v=Cghg-QyTP_M
Like what?
https://macleans.ca/facebook-instant-articles/boo-a-brief-hi...
Btw I think social media is a very dangerous tool being abused and exploited by many.
I assume that the current install process is fairly "human optimal". How much of that changes with your product to make it "automation" optimal? Were there small changes here for big wins? Or are you still using the same process with complex automation (the GM way), and do you see challenges arising from that?
One thing we're already doing that's different from human installers is the bay pre-assembly approach you can see in our demo video, where we build chunks of 5-9 modules on a tube and carry them out to the field. Ordinarily humans put up all of the tubes, then all of the brackets, then all of the modules. The pre-assembly approach allows us to do a bunch of work in a more controlled factory environment on a different part of the site.
Longer term there are other ideas we have, but since everything has to last for 30-40 years out in the field you have to be very strategic about making the big changes...
(Although, Wikipedia is telling me dishwashers already existed at the time of the Jetsons, so my analogy is bad, and I should feel badly).
Humanoid robot has much more Jack of all trades potential.
Originally "robot" was used only for humanoids, but the only essential capability that distinguishes robots from earlier machines are the manipulators, so in all other respects they may not resemble humans or animals nowadays.
Seems like you could fold them like an accordion into whatever max size can be carried by a forklift, drop them in place, pull them out to unfold, screw the whole prewired assembly down with ground screws, and plug it into the neighboring assembly?
This would probably work best with E/W racking, would go from tightly packed to corrugated.
I’m sure there are challenges I’m not considering, but with modules getting insanely cheap, it seems like racking/assembly/inverters are starting to dominate, and scaling up from single modules to full assemblies assembled in factories seems like a good next step.
(I’ve been building my own 20kw ground mount array, so I’ve had plenty of time to daydream about something we could’ve just dropped and pulled out)
A couple of ways this shows up: 1. making your system rugged to handle extreme weather. These sites need to last 30-40 years without breaking, which pushes you to use pretty heavy + large components. Piles get driven several feet into the ground, so even with unfolding systems you're still probably pile driving. 2. tracking -- adding the complexity to enable rows to rotate to face the sun makes it harder to make the system portable. Basically all large-scale solar getting built is single-axis tracked. 3. module cleaning -- it's important that vehicles can traverse rows to clean dirt off of modules on a regular basis. Having structure that connects the rows together may make this more difficult.
None of these are intractable on their own but together they make the approach you describe very difficult.
With panels getting so much cheaper, is tracking necessary still? Maybe if land is at a premium, but I know the aerocompact commercial deployments don’t support tracking.
Factory assembly and field deployment is a far faster, safer, and higher quality method of building solar farms.
Now that that’s taken care of, we just need some way to get the interconnection queues moving faster :-)
So ... tired ... of ... this.
This is NOT a shortage. This is lack of salary. The oil industry somehow manages just fine to get workers to really out of the way places.
Apparently the companies aren't drowning in enough projects to pay more money.
Why is that?
What's the most labor intensive part that remains when robots are putting the modules into place on racks? Attaching cables? I would guess that robots still aren't dexterous enough for that.
Still cheaper than a metal roof intermediate layer, though. But in places like Germany overhead glass outside of some limited applications (notably greenhouses for farming) isn't allowed to drop on anyone's head just from a sharp pebble impacting and cracking a glass layer. Sadly the regulations are just as strict for a home carport/garage-without-door as for a majestic train station hall. So aside from the worse cooling it's in practice cheaper to just use a corrugated sheet metal layer to catch rain and falling glass, with the structural support for both unified.
For traditional human installation, we're already approaching the limit of reasonable module sizes -- today's modules take two people to lift and align.
Are there installers that actually use humans to lift solar panels onto racking on solar field projects? I figured everyone was using vacuum pump panel lifters instead of humans these days as it’s much faster and less prone to injury.
One example: https://unimove.com/lift/solar-panels/
These can easily be fabbed up by a single person, it’s just some steel and a couple vacuum pumps.
Ahh - better - here is one at scale from a year ago. https://www.youtube.com/watch?v=72UlzbsON4M
West Texas:
Perhaps the next size up would be the interior dimensions of a cargo container so they could slide in/out like sticks of gum.
If the new robot can install panels very quickly, but you need 3 weeks to flatten the field to unreasonable tolerances or install tons of dedicated framing and supports for it to function, your efficiency gains or cost saving numbers will take a hit.
Those considerations are unsexy and boring, they put a shade to the press release, yet are very important to assess the value of the innovation. I'll trust a company when they start being transparent about these aspects.
These folks built a robotic restaurant. It took the form of a food truck in a shipping container, and they put it in our little food truck area here. I then saw for the next 3 months, engineers frantically running around a semi operational thing, until they took it away because it didn't work.
The investors could have just invested in some food trucks, and made money.
Launch HN: Charge Robotics (YC S21) - Robots that build solar farms - https://news.ycombinator.com/item?id=30780455 - March 2022 (81 comments)
pretty cool.
source : have a solar engineering firm in Kenya, and have never had problems with panel installation labor (QA is another story).
Now try to find ANYBODY willing to work out in the middle of the Mojave or Sonoran deserts. Espero que hables bien español.
It seemed like a pretty solid idea re: the QA part.
On the domestic side we need go make rooftop solar more efficient to install for countries like the UK, there aren't enough people to do it there either.
Broadly, I think there's so much potential for computing in construction. If we can model logistics decisions as optimization decisions, we can leverage incredibly efficient solvers to extract efficiencies. The challenge, of course, being the "info pipeline" - integrating machines with construction processes and planning software and orchestrating it all in a way that adds value.
If an entire construction plan is modeled, you could also compile it into a staged procurement/shipment plan, and essentially orchestrate the supply chain in sync with the jobsite activities.
It'd be cool to have more info on how the process looks like with/without these machines, how rates of specific activities are impacted etc. but a bunch of this stuff is probably proprietary.
This really is the "lowest hanging fruit" of the construction industry.
Did the tolerances truly need to be so tight? Probably not, and just set so by a designer somewhere. But the panels do have to line up for tying into conduit that goes to (pad mounted) transformers and distribution equipment, etc.
What I meant was, for something like a building, the plan will describe the invariants of the design, but what it really reveals are all the available variants in achieving that outcome on the actual site. With a solar farm that available variance is actually very small because the overall "plant" has very little additional or emergent structure.
This is somewhat like AVX - it only works if you have large enough vectors (a simple operation that is repeated a million times on contiguous data). Arbitrary buildings are much harder to optimize.
One of his greatest quotes is ‘We already have 3d printers for buildings, they’re called cranes’.
His advice every time I send him one these articles usually rounds to: ‘If these smart@$$ %|~>heads would spend a couple of weeks on the job actually sweating and doing some real labor, they’d realize the problem isn’t lifting things, it’s organizing it.’
For one thing, he's already identified but failed to realise the problem: organising labour to be on site and ready to work is expensive and tricky.
But when there's less labour and more machines, you're now only paying opportunity cost, not direct costs of people turning up and being unable to complete the job because it's not ready to go yet.
And unlike people, the quantity of capable machinery can constantly increase and thus the cost per unit decrease.
Eventually the machine is cheap enough to leave on site even if something else delays it.
The robot presumably doesn’t need many breaks and can go all night. But any outage can’t have someone else sub in.