In the half-light of a Chennai morning, where the air tastes faintly of diesel and ambition, a young man named Naveen Kumar stands at his desk and folds towels with unsettling precision. Not for a hotel, not for a spa, but for something far stranger: he is a tutor to the machines. A tiny cog in the booming global trade in physical AI, the industry determined to teach robots how humans actually move.
Welcome to the era of Arm Farms—vast, humming workshops where humans puppeteer machines, annotate their clumsy attempts at existence, and record their own bodily rituals for the coming mechanical generation. It’s all slightly uncanny, slightly comical, and just eerie enough to make your skin prickle.
After all, it’s one thing when AI learns our words. It’s quite another when it starts learning our wrists.
The New Gold Rush: Human Movement
Large language models have already inhaled the internet and learned to imitate our texts, our photos, our arguments. But physical reality—folding, gripping, manoeuvring a thin tea towel without flinging it across the room—is harder. More mysterious. More… human.
So companies around the world are rushing to bottle that mystery.
In the industrial town of Karur, Naveen straps a GoPro to his forehead and begins his ritual: pick up towel (right hand), shake (both hands), fold thrice (symmetry essential), place in the left corner (no sloppiness tolerated). Deviate by a single movement, a single second, and he must start again. It’s choreography as purgatory.
His company, Objectways, sends these videos to robotics firms abroad—200 towel-folding demos here, a batch of box-flattening movements there. Everything labelled. Everything boxed-in. Every gesture reduced to data points that can teach a robot how to behave.
It is human life turned into a textbook for mechanical children.
And if that’s not freaky… well, wait for what comes next.
The Rise of the Teleoperator
Many robotics firms now rely on human beings who operate robots from afar—joystick-pilots who guide metal limbs in dingy warehouses. East European barns full of operators controlling androids on the other side of the world. Factories where remote workers make robots pick up cups, make tea, plug cables, and tidy up tables.
Think of it as the world’s most tedious video game—except the stakes are real, and the players are training their replacements.
Critics call these machines sizzle over substance. Pretty to look at, impressive onstage, but hollow beneath the surface—marionettes with a wi-fi connection. Yet companies like Micro1, Deepen AI, and San José’s Figure AI are convinced: with enough examples, enough manoeuvres, enough labelled towel-folds, the robots will eventually learn to act without their puppeteers.
That’s the dream. Or the nightmare, depending on where you stand.
The Smart-Glasses Harvest
If teleoperation is too clumsy, there’s another method: capture absolutely everything humans do, all day long.
Workers in Brazil, India, Argentina, America now strap on smart glasses that record each flick, tap, chop, reach and stumble. Entire libraries of mundane bodily movements—opening fridges, brushing crumbs, stirring tea—becoming the source material for the robot apprenticeship.
Figure AI has even partnered with a real estate giant to collect footage inside 100,000 homes. Meta-backed Scale AI has already gathered 100,000 hours of similar film.
It’s domestic surveillance repackaged as innovation. A quietly revolutionary—and quietly unsettling—idea.
When Clients Want Purple Tables
Not all of this runs smoothly. In Bengaluru, a young entrepreneur tried to capitalise on India’s inexpensive labour by offering physical data collection on demand. He quickly discovered the truth: robot trainers are picky. Maniacally picky.
Clients didn’t just want demos of robotic arms cooking—they wanted the exact same robotic arm, operating on a table of a precisely specified colour, under purple lighting. A choreography so specific that even cheap labour couldn’t make it profitable.
Humans have preferences. Robots require obsessions.
And Back in Karur…
Objectways isn’t slowing down. It has begun capturing footage of humanoid robots themselves—helping them learn to fold towels without throwing them across the room in a tantrum of servos.
Sometimes the robot flings clothes into the void. Sometimes it sweeps the entire pile onto the floor like a bored teenager. But it is learning.
And the towel-folders of Karur know exactly what they’re witnessing. As one employee put it, with unsettling clarity:
“In five or 10 years, they’ll be able to do all the jobs, and there will be none left for us.”
Spoken lightly. But with an undertone thick as storm clouds.
The Freaky Future of Bodies and Machines
The rise of Arm Farms reveals a strange truth: AI doesn’t just want our data; it wants our gestures, our grip strength, our domestic choreography. It wants the tacit knowledge our bodies carry—the quiet, unnoticed intelligence in our wrists and elbows.
We’ve now reached the point where humans are training robots to replace the very things that make us human: the ability to move through the world with grace, clumsiness, curiosity, indecision, flair.
Perhaps the strangest part isn’t that robots are learning from us. It’s that we’re teaching them with such devotion, precision, and discipline—folding towels again and again under the neon lights, documenting our own physical poetry for a future that may not require us.
The machines are learning. The humans are sweating. And somewhere in a warehouse, a robot raises its arm, reaches for a towel, and folds it just slightly better than yesterday.
A gesture almost human.
Almost.
Robert Hubert: The Man Hanged for Starting the Great Fire of London