Skip to content
‹ Back to IndustriesEngineering & Construction

MIT voxel blocks let imprecise robots build load-bearing demos

MIT visiting researcher Miana Smith’s voxel blocks are designed to absorb about 10 millimeters of placement error so low-cost robots can assemble room-scale, load-bearing demos. The first field partner is Druk Holding and Investments for a planned outdoor test tied to housing at Gelephu Mindfulness City, and Smith is targeting concrete formwork next.

This story was produced through MarketScale. See how Engineering & Construction teams put it to work with Partner & Channel Enablement.

By MarketScale Newsroom · Miana SmithMit Center for Bits and AtomsVoxel ConstructionConstruction Robotics
Share
MIT voxel blocks let imprecise robots build load-bearing demos

Key takeaways

01

Putting the accuracy in the block lets a robot miss by about 10 millimeters and still assemble load-bearing demos; that tolerance is about 100× looser than typical industrial arms (an illustrative ratio from the reported figures, not a measured claim).

02

Cheap robots shift the hard part into the component: the voxel approach depends on producing complex lattice blocks economically, and Smith’s focus on automating their production suggests she sees that as the key hurdle.

Free workspace

Turn your Engineering & Construction expertise into content.

Record interviews, organize footage, and write with AI on a free trial of the MarketScale platform for qualifying companies. No demo required, no credit card.

Try it Free

On most factory floors, a robot that misses its mark by 10 millimeters gets pulled for recalibration. Industrial arms typically hold tolerances near 100 microns. In Miana Smith's lab at MIT's Center for Bits and Atoms, a 10-millimeter miss is just what the robots do, and the structures still go up load-bearing.

MIT News profiled the work on November 18. Smith, who has three MIT degrees (’21, SM ’23, PhD ’26), finished her PhD this year and stays on as a visiting researcher with the Center for Bits and Atoms team in MIT's Program in Media Arts and Sciences. She designs lattice blocks, which her lab calls voxels, that snap together and absorb placement error as the structure grows. The profile notes that a 10-millimeter margin would look like a failure to most engineers. Here it lets her robots cost a few hundred dollars apiece instead of tens of thousands.

Set the two reported tolerances side by side and the gap is about a factor of 100: 10 millimeters is 10,000 microns, against roughly 100 microns for a typical industrial arm. That ratio is this article's own arithmetic on the two figures MIT News reported. It's a way to picture the gap, not a measured performance claim.

For anyone at a contractor or developer weighing construction automation, the work changes which question comes first. Instead of asking how accurate a machine is, ask how much error the parts can tolerate.

The precision tax on efficient buildings

Engineers already know how to use less material. The MIT News piece, written by Sydney Campbell of MIT's Office of Graduate Education, points to steel-framed buildings, where an estimated 30 to 40 percent of the steel could come out without weakening the frame if beam cross sections were customized. It rarely happens. Custom parts need one-off design and engineering, the manufacturing is often impractically expensive, and installing them on site takes highly skilled labor. So oversized, uniform members win, because they're cheaper to make and assemble at scale.

Digital fabrication was meant to narrow that gap. With 3D printing and robotic assembly, a complex design no longer has to cost more to produce. Those tools carry their own costs, though. Placing framing members or extruding concrete over a 10-meter span frequently calls for a machine with accuracy of a few millimeters, and equipment that precise is difficult to build and costly to purchase.

Look at it that way and the extra steel in a typical frame comes from manufacturing and labor, since the design knowledge already exists. Any fix has to make complexity cheap to build. Making it cheap to draw doesn't help.

Where the precision budget sits

About 100 microns
Typical tolerance of an industrial robot, the accuracy most automated assembly is designed around
About 10 mm
Margin of error Smith's voxel-assembling robots work with
A few mm over 10 m
Accuracy conventional digital fabrication often needs to print concrete or place framing on a building-scale span
30 to 40%
Share of steel that could be removed from steel-framed buildings with customized beam cross sections

MIT News

How a block corrects its robot

Smith reverses the typical arrangement. Conventional systems combine simple material with a precise machine. Hers combines complex material with a simple, imprecise machine. She likens it to Lego. A child builds very imprecisely, but the finished Lego model still comes out precise. As she told MIT News, that is because the bricks are designed with clever mechanics that stop errors from adding up as the structure gets bigger.

Getting there took a structural change. For a long time her team built with small units that each needed as many as 24 connectors. Every joint was another chance for weakness or a failed placement, and that kept robots from building anything large. The fix, which Smith has described as almost embarrassingly obvious, was to make the blocks bigger and attach as many internal joints as possible before a robot ever picks one up. She calls the result a structurally active error-correction feature, the thing that lets robots build big and still carry load.

The lattice changed too. The team moved to a more heavily interconnected, stretch-dominated geometry, which Smith says beats a bending-dominated lattice on weight-to-stiffness and weight-to-strength. One joint design now works across plywood, bio-based plastics, aluminum and steel. The plywood version pairs Baltic birch with 3D-printed, compostable bio-based plastic, which Smith describes as a low-embodied-carbon way to make structural units from fairly standard construction materials.

According to MIT News, changing the material, the fabrication process and the lattice density within that common joint produces voxels with different properties. The outlet adds that since the system does not depend on any single material, architectural ambition need not be sacrificed. For a designer, this suggests material could become a project-by-project choice within one assembly system, rather than a reason to change systems.

The lattice geometry behind the voxel is not new. What Smith's research adds is a way to automate making and assembling the blocks so they can work at scale. That automation has been her main research focus.

Why many cheap robots can beat one precise one

Cheap robots bring a second benefit that matters on a schedule. A few hundred dollars per unit means many robots can work in parallel, where one expensive, precise machine would do everything slowly. The block carries the accuracy, so the robot can miss by about 10 millimeters and still build something load-bearing.

Smith's route to that idea ran through some unlikely places. In high school she rebuilt motorcycle engines through her school's STEM program and did electrical engineering work at a plasma physics lab at UCLA. While an MIT mechanical engineering undergraduate, she took a research position with architecture professor Larry Sass that introduced her to digital fabrication for buildings. Then came class 4.140, How to Make (almost) Anything, taught by Center for Bits and Atoms director Neil Gershenfeld, later her PhD advisor. There she became captivated by the notion that the whole world, buildings among it, might be assembled from an engineering version of Lego bricks.

Smith spent two years working with living systems in another Media Lab group before she moved into robotics. In that work she engineered bacteria into woven textile fibers, and she sent honeybees into microgravity to see whether they could survive. They did. She says that living organisms are, at their core, distributed assembly systems as well. By her account, that realization is what led her from small bugs to robots.

That background helps explain where the design puts its effort. In a distributed assembly system the accuracy has to live in the parts, so adding agents adds capacity without adding precision. For a builder, it suggests throughput would scale by adding inexpensive units rather than by buying a faster, more accurate machine. That's a different capital decision from the one industrial automation usually presents.

The strongest objection is that the cost doesn't disappear. It moves. A complex block has to be manufactured, and the economics hold only if voxels can be produced as cheaply as the robots that place them. Smith's own focus on automating production suggests she sees that as the real problem to solve, and it's the line item a buyer would need to see priced before anything else.

Checking the 80 percent claim

Smith estimates that buildings made this way could cut embodied carbon by roughly 80 percent compared with standard concrete construction. The baseline is big. MIT News notes construction is often credited with more than a third of global greenhouse gas emissions, mostly because buildings use more material than they need.

Other tallies land in the same range. The Global Status Report figures compiled in Wikipedia's green building entry put buildings at 37 percent of global energy use and energy-related CO2 emissions as of 2020, and at 39 percent of global CO2 emissions once the manufacturing of building materials is counted. Material production is the part of that total the voxel approach targets. So the comparison point Smith chose, standard concrete construction, matters as much as the percentage itself.

Treat the 80 percent as a researcher's estimate, not a measured result from a finished building. So far the demonstrations are room-scale plywood structures, and outdoor testing is still ahead. Teams setting embodied-carbon targets can watch the figure, but it isn't ready for a procurement spec.

From Bhutan housing to concrete formwork

The first field partner is Druk Holding and Investments, a major Bhutanese company focused on building a sustainable economy. The work is for its next housing development at Gelephu Mindfulness City. Bhutanese officials wanted construction methods that use local materials and can be reconfigured over time, which Smith saw as a natural fit for the voxel work. Her team has built room-scale, load-bearing demonstrations from plywood blocks, including a small house, and plans to test the system outdoors at a similar scale.

The commercial target is concrete. Smith hopes the same reconfigurable block logic can be used for formwork, so concrete can be shaped more efficiently. Complex molds are very expensive, and she expects a reconfigurable system to bring that expense down by a wide margin. Smith says she wants to know that her work makes a difference. In her view, everyone in her generation, at the very least, regards the climate crisis as the doom that lies ahead.

She's starting a company around sustainable concrete construction through an Activate Fellowship, a science entrepreneurship program that helps researchers commercialize their work. Gershenfeld has said he didn't first see her as an entrepreneur but now considers her the best kind, driven by the change she wants to make more than by a business plan.

Formwork is where general contractors and concrete subcontractors would feel this first. That's especially true for those whose margins on complex geometry depend on custom molds that get used once and thrown away. A reusable, robot-assembled mold system would change the cost of a curved wall or a variable cross section without asking anyone to give up concrete.

When a voxel-based formwork product reaches the market, ask three things: the cost per block at volume, how many pours a block set survives before reconfiguring, and what placement tolerance the field robots actually hold on site.

The research group isn't standing still either. According to the full MIT News profile, a new cohort of students is already exploring robots of different shapes working together on bigger builds. The next concrete marker is the outdoor test at Gelephu. A plywood house that stays load-bearing through Bhutan's weather would say more about the blocks than any lab demonstration has so far.

Featured companies

Your experts belong here

Every story in MarketScale Engineering & Construction starts with a company putting its project engineers, superintendents, and estimators on the record. Buyers are already reading this topic. The only question is whose experts they find.

Owners shortlist firms they already trust, and your field leaders become the reason your name is on that list.

Book DemoSee how it works15 minutes, straight to a calendar.

About the author

MarketScale Newsroom
MarketScale NewsroomEditorial Team, MarketScale

The MarketScale Newsroom reports on the companies, technologies, and trends shaping 16 B2B industries. It turns primary sources and expert commentary into clear, useful coverage for the people doing the work.

B2B Weekly

The week in Engineering & Construction, and sixteen other industries, every Monday.

Ten stories, one-line takes, five minutes. Free.

Engineering & Construction: are you visible to AI?

Before they reach out, Engineering & Construction buyers ask AI engines which vendors to trust. Explore how your experts, customers, and partners can become useful content for buyers and AI search.

Free Trial

You just read one Engineering & Construction expert. Your company is full of them.

This article was produced through MarketScale. The same platform turns your project engineers, superintendents, and estimators into the articles, video, and social content Engineering & Construction buyers are searching for. Start a free trial and see it with your own people. For qualifying companies, no credit card, no demo required.

NPS +73 · 1,000+ creators · 38+ countries

What your free trial includes

Hands-on access to the MarketScale platform
Media requests to your crowd, remote recording, AI writing tools
No demo required. No credit card.
For qualifying companies. Company confirmation required.

More Engineering & Construction Insights

Why a modular data center's 7.5-month head start depends on grid power

Why a modular data center's 7.5-month head start depends on grid power

Prefabricated modular construction can cut a 50 MW liquid-cooled data center build in Asia Pacific from about 22 months to about 14.5, depending on the market. But grid and equipment lead times can still set go-live dates. Where power is already secured, Turner & Townsend says the time saved can convert into earlier revenue; where it isn’t, it can let owners delay construction commitment closer to energization. The tradeoff is risk shifting toward factory slots, shipping, logistics and on-site interfaces.

  • 01Where the grid date is still uncertain, a 7.5-month modular saving buys the ability to commit construction money closer to energization, not earlier revenue.
  • 02Licensed electrical work is the scarcest trade in several Asia Pacific markets, and moving it into a factory can cut that on-site requirement by close to 80%.
  • 03Before signing a fixed-price module package, test the factory slot, shipping window, duty and integration assumptions behind the ready-for-service date.

Nov 21, 2026

ArcGIS Pro's November update brings OpenRoads corridor geometry into GIS

ArcGIS Pro's November update brings OpenRoads corridor geometry into GIS

Esri's November ArcGIS Pro release brings Bentley OpenRoads Designer corridor geometry and TIN surfaces into GIS. It also lets analysts propagate Building Layer settings for more consistent setup and adds Fine, Medium and Coarse graphic detail options for Revit models.

  • 01ArcGIS Pro now reads alignments, profiles, station equations and TINs from OpenRoads Designer, letting GIS teams combine corridor design data with Revit structures and Civil 3D site work in a single map
  • 02Preparation work for BIM-to-GIS conversion moves upstream to model setup: choices about graphic detail, fields, definition queries and coordinates now land with whoever configures the design before Esri does the conversion
  • 03GeoBIM’s new “3D ArcGIS” view workflow filters Revit content at the source, and ArcGIS Enterprise 12.2 adds a pre-sync analysis report to check georeferencing and coordinate system details, flag missing information and link to resources for fixes.

Nov 21, 2026

206 construction AI studies show far more work on emissions than on extreme weather

A review of 206 construction AI studies spanning a decade finds AI can support climate change mitigation and adaptation across every project stage, but the evidence is lopsided. Tools that run assets more efficiently are rated relatively mature, while generative design for climate resilience and fully integrated digital twins are still largely in pilots.

  • 01After 206 studies spanning a decade, the construction AI rated relatively mature is mostly about running assets more efficiently, while AI that would design buildings to ride out extreme weather is still largely in pilots.
  • 02Any AI pitched as a climate win should be judged net of the compute and electricity it consumes to train and run.

Nov 20, 2026

Explore More Engineering & Construction Insights

Read more expert perspectives from across Engineering & Construction.

Browse Engineering & Construction Hub

About the Expert

MarketScale Newsroom
MarketScale Newsroom

Editorial Team

MarketScale

The MarketScale Newsroom reports on the companies, technologies, and trends shaping 16 B2B industries. It turns primary sources and expert commentary into clear, useful coverage for the people doing the work.

For B2B teams

Your experts could be publishing here

Stories like this one run on content MarketScale captures from real practitioners. See how your team's expertise becomes coverage in Engineering & Construction and beyond.

Book a Demo

Or call us. No forms required. We pick up. 214-945-2512