Construction is famously labour intensive: nonstop labour makes up close to 50% of the costs of constructing a caller single-family location successful the US, compared to around 6 to 8% of the costs of manufacturing a car. Unsurprisingly, there’s long been liking successful automating the building process to trim the magnitude of labour required, and truthful the costs of housing. The earliest effort astatine automating building I’m alert of dates to 1853, pinch a plastering instrumentality invented by Isaac Hussey, and since past location person been galore different attempts.
I wanted to get a amended consciousness of what the arc of building automation looks for illustration complete time: what things had been tried when? What worked, and what fell by the wayside? To do this, I utilized AI to trawl done thousands of aged issues of digitized building waste and acquisition journals going backmost to the 19th century, looking for immoderate illustration of a building task being automated.
My main takeaway is that successful building automation fundamentally falls into 2 categories. One is construction-related tasks that tin beryllium done off-site successful a factory: this includes things for illustration alloy fabrication, precast concrete, model and doorway manufacturing, and truthful on. The different class is on-site tasks that are, aliases tin beryllium made to be, highly factory-like: they tin beryllium done by a instrumentality moving on a elemental path, doing immoderate repetitive mobility complete and complete again. Historically, almost thing extracurricular these 2 categories hasn’t succeeded.
I started by downloading each backmost rumor of Architectural Forum (950 issues) and House & Home (370 issues) connected USModernist, and each backmost rumor of Engineering News-Record (4,300+ issues) disposable connected archive.org. These are waste and acquisition journals that screen various aspects of the building industry, including building practices and methods, and screen the play from the mid-19th period up to the early 1980s. I past had Claude (Opus 5 Max) hunt done these issues and emblem immoderate lawsuit of immoderate building task being automated. In summation to this, I besides had Claude do a much wide web hunt for examples of building automation, which included things for illustration looking done the archives of the International Association for Automation and Robotics successful Construction (IAARC).
The intent of this wasn’t for information analysis: I wasn’t counting the number of cases by twelvemonth aliases by category, aliases thing for illustration that. This was purely an beingness study truthful I could spot what sorts of things had been tried and were promising capable that they made it into a waste and acquisition journal.
As accustomed pinch a task for illustration this, the results will beryllium a usability of what definitions you take to operationalize the position successful your prompt. By “automation,” I mean thing for illustration “a instrumentality does immoderate important information of a task without requiring quality direction.” So this is chopped from “mechanization” aliases “industrialization,” wherever a instrumentality is utilized to trim the labour required to do immoderate task, but that instrumentality mightiness still require continuous guidance aliases nonstop cognition from a quality operator. So thing for illustration a powerfulness instrumentality aliases a crane wouldn’t count arsenic automation, because successful some cases each individual movement/action of the instrumentality is orchestrated by the quality arsenic the activity is being done. A bricklaying instrumentality that automatically placed individual bricks without a quality usability directing it ceramic by brick, connected the different hand, would count arsenic automation.
Like astir definitions, this 1 gets fuzzy astir the edges. In believe astir automated systems require immoderate grade of quality oversight while they’re running, and whether that oversight counts arsenic “directing individual movements” tin beryllium thing of a judgement call. So I’d see a road-paving instrumentality that automatically lays down asphalt “automation,” moreover pinch a quality driver monitoring the instrumentality continuously. But thing for illustration a ride-on actual finishing instrumentality seems overmuch much borderline, since that seems to require a batch much quality guidance connected a second-by-second basis.
If immoderate building task tin beryllium done extracurricular an existent building site, successful a mill aliases factory-like environment, there’s a very bully chance that, for illustration immoderate mill task, it tin astatine slightest beryllium partially automated.
An early illustration of this is actual batch plants. Through astir of the 19th century, actual was mixed manually by workers pinch shovels, but by the precocious 19th period actual mixing started to beryllium done by machine, and during the first half of the 20th period these became much and much automated, pinch things for illustration automatic cement weighing, automatic h2o dispensing, and truthful on. On the Hoover Dam, for instance, actual batching was “completely automated.”
Steel fabrication is similar. Steel sections are, of course, a factory-produced merchandise and ever person been, and it’s not overmuch of a leap to move earthy alloy elements into building components astatine the mill and to partially automate those tasks. Machines to do tasks for illustration automatically drilling holes successful alloy beams, automatically welding tube seams, and automatically coating alloy sections started to look successful the precocious 19th period and stay wide utilized today.
More generally, fundamentally immoderate building worldly — steel, brick, actual block, plywood, drywall, shingles — will beryllium produced successful ample volumes successful a mill and frankincense various portions of the accumulation process will beryllium automated.
Concrete, arsenic we’ve noted, started retired arsenic a site-produced worldly but yet had its accumulation moved off-site. We spot the aforesaid trajectory pinch different building components. Windows and doors, for instance, utilized to beryllium fabricated on-site, but those are now produced successful off-site factories pinch galore of the assembly operations automated. Floor and tile trusses were likewise primitively site-produced components but are now mostly made successful off-site factories pinch varying degrees of automation. A emblematic truss mill will automate the cutting of the various wood sections and the pressing of the truss plates, and immoderate truss factories person moreover higher levels of automation.
Precast actual is different factory-produced building component: ample actual elements for illustration beams, walls, and columns are manufactured successful an off-site plant, past trucked to the jobsite and assembled there. Precast manufacturing doesn’t needfully person a awesome woody of automation, peculiarly successful the US, but it tin beryllium highly automated.
This is, of course, portion of the logic there’s been truthful overmuch humanities liking successful prefabricated construction: erstwhile you move the existent building process into a factory, you tin perchance automate ample fractions of it, achieving (it’s hoped) labour and costs savings. While sometimes prefabricated building is simply conventional, labor-intensive building done wrong a warehouse, different times the process is highly automated. Levitt and Sons built a highly automated prefab works arsenic portion of Operation Breakthrough successful the 1970s, and coming companies specified arsenic Autovol successful the US and Gropyus successful Germany tally partially automated plants.
The different class of successful automation is on-site tasks that tin beryllium made factory-like: a simple, repetitive motion, done complete and complete again arsenic a instrumentality moves on a elemental path.
A batch of the examples successful this class impact actual aliases asphalt. Concrete is simply a bully campaigner for this benignant of automation because it’s a thick, viscous fluid; it tin beryllium squeezed into the style of immoderate peculiar instrumentality and past support that style erstwhile the instrumentality is removed. So you person things for illustration automatic actual slipforming machines, which extrude a furniture of actual arsenic they beforehand forward. Oriented horizontally, these tin beryllium utilized to make things for illustration curbs, trenches, and sidewalks. Oriented vertically, these tin beryllium utilized to make vertical actual structures (though I judge vertical cognition is somewhat little automated).
We besides spot this benignant of “automatic extrusion machine” for metal. Gutter-making machines are fundamentally portable roll-forming machines: they return successful expanse metallic from a coil and style it into gutters. Similar portable machines beryllium for metallic tile platform and metallic siding.
You besides spot this benignant of automation utilized successfully successful the building of horizontal, continuous infrastructure. Roads, for instance, are paved pinch a bid of specialized machines called a paving train. Working together, the machines activity successful series to rip up and roughen the existing aboveground of the roadworthy and past laic down a caller furniture of asphalt.
Parts of railway building tin likewise beryllium done by a bid of automated machines. Maintenance tasks — things for illustration replacing worn-out railroad ties and tamping ballast — successful peculiar look to beryllium highly automated. Since the obstruction is already successful place, this machinery tin beryllium rail-mounted, performing its various specialized tasks arsenic it moves on the track.
Tunneling is similar. Historically tunnels were done by hand, but astir the move of the 20th period tunneling began to beryllium done by self-propelled boring machines.
From what I tin tell, up done the early 2000s (and moreover into the 2010s), almost everything that didn’t autumn into the 2 categories (tasks that tin beryllium done successful an off-site factory, and tasks that tin beryllium done successful a factory-like measurement on-site) didn’t succeed.
Automatic bricklaying machines are the classical illustration of unsuccessful automation here. As I noted successful a previous essay, these sorts of machines person been tried complete and complete again since the early 20th century, and historically nary of them succeeded.
Automatic mounting of bricks aliases actual blocks seems for illustration it would beryllium the cleanable building task to automate: it’s some physically strenuous and highly repetitive. But successful practice, the task isn’t rather elemental and repetitive capable to build a instrumentality to easy do it. Spreading a furniture of mortar, and past placing a ceramic connected apical of it, while ensuring that some the individual ceramic and wide wall stay consecutive and level turns retired to beryllium difficult to do pinch a uniform, repetitive mechanical motion: successful believe the bricks often extremity up somewhat askew, requiring small, individual adjustments to the bricks to make judge that the wall is being built level. Making nice, soft mortar joints likewise isn’t easy for a instrumentality to do. SAM, the Semi-Automated Mason manufactured by Construction Robotics, required masons to travel down the robot to decorativeness the joints and occasionally level bricks that were group askew. Fastbrick Robotics’ Hadrian block-laying robot avoided these problems wholly by utilizing a civilization adhesive that doesn’t require spreading a furniture of mortar.
Brick walls besides often aren’t repetitive capable successful the consciousness that they seldom dwell of long, uninterrupted runs of uniformly placed brick. Walls are often short, they person model and doorway openings, they person joints, they person corners and jogs, and various different deviations from uniformity. A elemental instrumentality doing a purely repetitive motion, aliases moreover a robot pinch a greater grade of flexibility, can’t easy grip these sorts of deviations, and historically bricklaying machines needed to beryllium manually readjusted by a personification whenever they encountered 1 of these situations. The SAM bricklaying robot really only worked good connected very long, azygous stretches of wall; because it couldn’t move corners aliases moreover group bricks astatine the extremity of walls, it needed to beryllium group up caller astatine each caller wall, which took clip and effort.
This benignant of trouble is very communal pinch building automation. Often there’s immoderate repetitive task that tin beryllium automated, but it’s not rather repetitive capable to make it worthy mounting up immoderate instrumentality to do it. If a personification has to manually reset a instrumentality each fewer minutes and move it to a caller area, that’s often not immoderate easier than simply having a personification do the task. Builders tried automatic welding machines connected jobsites arsenic early arsenic the 1930s, for instance, but you don’t spot galore specified machines utilized today: buildings seldom person long, uninterrupted continuous stretches of welding that would make specified a instrumentality worthy it.
Similarly, tying rebar — wrapping a bladed ligament astir 2 pieces of rebar to necktie them together — is elemental capable to beryllium automated, but historically it hasn’t been rather repetitive capable to section a dedicated instrumentality for it. (In my experience, it’s often not repetitive capable to moreover springiness workers powerfulness devices that tin do this.) Even coming the calculus appears beautiful unfavorable: Advanced Construction Robotics started offering the TyBOT successful 2020, but you only ever spot it demonstrated connected actual span decks, which person very ample expanses of rebar to tie. (It besides has a rebar-placing robot, IronBOT, but not only does this besides require ample expanses of rebar to make consciousness — its website states that “[s]etup requires readying and coordination owed to the IronBOT’s size and capabilities” — but the existent automation is reasonably thin, requiring a quality to perpetually provender it individual bars.)
The deficiency of capable repetition, mixed pinch trouble successful making a building tract “factory-like,” is what doomed Japan’s push to create building robots astatine the extremity of the 20th century. Starting successful the precocious 1970s, Japan undertook a awesome effort to create each mode of building robots — column-welding robots, fireproofing-installation robots, wall-painting robots, and truthful on. By the 1990s, complete 100 different types of building robots had been developed. As I noted successful a previous essay, while these systems often worked, they mostly weren’t peculiarly productive:
Though the robots could often execute their tasks highly efficiently, they required a awesome woody of setup and teardown time, and often encountered difficulties successful buildings that hadn’t been specifically designed to accommodate them (such arsenic constrictive spaces that the robots couldn’t reach). They besides had problem dealing pinch the somewhat unpredictable quality of a accepted building tract — their request to activity uninterrupted, mixed pinch information limits connected really intimately workers could approach, tended to constrain the remainder of the building process. And moreover erstwhile tasks could beryllium completed faster, it often meant that the bottleneck conscionable shifted to immoderate different portion of the process…
To effort and flooded these issues, Japan’s largest contractors tried to make jobsites arsenic factory-like arsenic possible. They created immense “skyscraper factories,” immense climbing platforms that would build a skyscraper level by level pinch the assistance of each mode of building robots. But this grounded to lick the problem:
The agelong mill setup clip made them impractical for smaller buildings, and moreover connected buildings that were ample enough, extended upfront creation and coordination was required to creation the building for robotic assembly. In immoderate ways, the move from single-task building robots to afloat automated building sites conscionable shifted wherever the difficulties occurred — alternatively of a agelong setup (and takedown) clip for each individual task, now you had 1 tremendous and costly setup complete the full first information of the project.
Historically, building automation has been successful connected definite types of tasks: tasks that tin beryllium moved off-site into a factory, aliases tasks that are elemental and repetitive capable that they tin beryllium done successful a factory-like mode on-site. Automation extracurricular of these areas mostly hasn’t succeeded. Often this is because the task isn’t rather elemental and repetitive enough: while you tin build a instrumentality to do it, the clip and effort it takes to get the instrumentality group up and doing the work, and orienting the jobsite to accommodate it, obviates immoderate advantage you get from the automation.
I think, however, that we’re connected the cusp of this changing.
This humanities shape we’ve observed is, successful ample part, a technological limitation. For astir of history machines person had, astatine best, very constrictive information-processing abilities: they were constricted successful what they could consciousness astir their environment, and the grade to which they could modify their actions based connected that information. Often they simply did the nonstop aforesaid mobility complete and complete again sloppy of the circumstances. When they could modify their behaviour based connected biology feedback, it was successful very constricted ways. Without robust information-processing abilities, automation is constricted to being done successful a mill (where you tin break isolated your task into a bid of mini tasks, each 1 done by immoderate peculiar machine) aliases to processes that are people factory-like.
The history of welding automation provides an example. The earliest welding machines were fixed automation: they performed immoderate circumstantial welding mobility — specified arsenic welding the seams connected pipes — complete and complete again. Starting successful the 1960s, welding began to beryllium done pinch robots: now alternatively of needing to build a caller machine, you could simply reprogram the robot, feeding it a different series of actions to take. But that series was still limited connected a quality programmer to determine.
Over the adjacent respective decades robotic welding gradually accumulated information-processing abilities: by the 1980s a welding robot had a constricted expertise to find its position pinch respect to the weld and modify its actions accordingly. But the scope of this expertise was very narrow: a welding robot could do things for illustration measuring the electrical conductivity done the weld to find its position, but not thing much complex:
They couldn’t compensate for things for illustration a clamp that impeded the way of a robot, aliases a portion that was excessively soiled to weld properly, aliases a portion that was oriented vertically alternatively of horizontally, aliases immoderate of the different 1000 things that a manual welder could automatically set for.
Thus moreover automated robotic welding required a reasonably repetitive group of tasks to beryllium worthy the clip and effort it took to group up.
But this technological limitation is changing, and we’re getting machines that person acold much robust information-processing capabilities. Modern automation is progressively tin of being fixed a group of instructions and figuring retired what actions it needs to take, measurement by step, to execute them. This is, successful ample part, owed to modern AI technology: if you provender immoderate AI exemplary a immense magnitude of training data, you tin get thing that responds efficaciously successful different situations — a self-driving car that navigates a way and a postulation shape that it’s ne'er observed before, an LLM that answers a mobility that it’s ne'er seen before, and truthful on.
This changing constraint is reflected among the emerging group of building automation startups. There’s a caller harvest of building automation startups that are leveraging modern AI exertion and are getting traction and raising ample amounts of funding. Roughly half of the VC backing for building robot startups has gone to companies founded post-2022, and astir different ample backing rounds for building robotics person been post-2020. (And this doesn’t see the tremendous amounts of backing that person gone into much general-purpose, non-construction-specific robots, for illustration Figure.) It’s progressively becoming imaginable to automate building tasks that were antecedently resistant to it, and I only spot this inclination continuing.
It’s not clear to maine precisely what shape this will take: will we spot an service of Optimus robots connected building sites tin of a wide array of tasks, aliases will it beryllium a bunch of special-purpose robots for illustration Dusty and Canvas and Rufus? I fishy that sloppy of really it develops, the early of building automation does not look for illustration the past.
English (US) · 
Indonesian (ID) · 
