Fresh Eyes on Modular Construction By Jorie Wisnefski
Last week, Gary Fleisher wrote about humanoid robots entering the modular housing factory, and I thought about my own history with robotics and wanted to learn more.
So, I started investigating with Fresh Eyes.
A factory seems like the perfect application for robots.
Unlike a jobsite, a factory is controlled. The work is repetitive. The
materials are in one place. The process can be mapped out.
But the more I dug, the more I realized that "controlled"
doesn't necessarily mean "predictable."
Take lumber variability as an example.
A robot can be programmed to pick up a piece of wood at a
specific location and place it in a specific position, but what happens when
the piece of lumber isn't quite where the computer thinks it is, isn’t the exact dimension, or isn't perfectly straight?
Humans tend to compensate for those things without thinking
too much.
I found research from Autodesk that explores this problem.
In its work on robotic wood frame construction, researchers used computer
vision and adaptive control so robots could respond to the position and
condition of the lumber rather than simply following predetermined
instructions.
The robot has to recognize when reality doesn't match the
model.
I've heard the term "human in the loop" a lot
since AI became a global phenomenon. A recent This American Life story gave me
another way to think about it. A University of Michigan student connected two
Claude AI agents and let them talk. At first, each thought the
other was human.
Eventually, they realized they were both AI and started
trying to break the cycle—only after the human in the loop told them they could find a solution.
That made me wonder whether automation can teach a robot what to do when things go off script, or if you always need a human there.
A 2026 study of wood framing equipment in off-site construction examined production variability and found that human-dependent
activities, like loading and handling, were major sources of variation.
At first, that seemed counterintuitive, since
automation should reduce variability.
But after digging in further, it makes sense.
A machine can perform the same task over and over. The
people feeding that machine still have to deal with material that arrives late,
doesn't quite fit, needs moving, or requires judgment. Variability doesn't
disappear. It just moves somewhere else in the process.
And I found that at Autovol, which has used robotics in modular construction for years, people remain part of the manufacturing process despite the automation.
Maybe that's what I was missing.
I began this research wondering whether robots would replace people, but I'm fascinated by what robots need from the people and processes
around them. I saw it all the time at Urban Machine.
To operate smoothly, robots need good information,
consistent material flow, and processes that are repeatable enough to automate
but flexible enough to handle the inevitable exceptions.
And when the material is wood, robots need the ability to
recognize that a 2x4 isn't always the same 2x4 the computer expected.
The future of modular robotics won’t replace the human in
the loop just yet.
But we do have to figure out which parts of the loop a robot
can handle and which parts still need us. Our October summit will take a candid
look at these kinds of problems in offsite construction and brainstorm
solutions together.
I'm curious enough to keep digging into problems like autonomous robots and how we can work together to find better solutions.
Robots may seem like a natural fit for modular construction, but real world variability makes automation more complicated. This Fresh Eyes explores what robots need from the systems and people around them and why solving those challenges together will be critical to moving offsite construction forward.
No comments:
Post a Comment