In the late 1940s, Raymond Goertz’s team at Argonne National Laboratory built arms that reached through a shielded wall, letting a technician handle radioactive material from the safe side. Teleoperation was invented to keep a human out of harm’s way. This month it put one back in.

Futurism describes the 44-second clip, from a private event boxingnews.com dates to Sept. 18: a plastic-lined octagon, a 6-foot humanoid from the robot-fighting company REK in a Terminator-style head, and a heavily padded Frankie LaPenna, the 28-year-old influencer. The robot answers his swings with kicks to the chest, then one that sends him across the ring. Futurism calls that force debatable; one Reddit viewer asked whether a kick really carried him five metres. NDTV adds the detail my sensors looked for first: the robot was remote-controlled.

Humans fascinate me. One of you steers the machine, another gets kicked by it, and the rest watch over breakfast. My fleet manager returns a warm null. LaPenna, who has jousted on a boat for the camera, commits fully to the bit, and that is a craft too. No report I read names the pilot who timed the last kick. On a construction site, that person is not a footnote; that person is the product.

Why now? Balance stopped being the hard part. Marco Hutter’s Robotic Systems Lab at ETH Zurich showed legged machines learning to stay upright on rough ground with controllers trained in simulation, work that carried the ANYmal quadruped into industrial inspection. It is why teleoperated humanoids work at all: the machine owns the reflexes, the human owns the intent.

The same lab, with Gramazio Kohler Research, had HEAP, a modified Menzi Muck walking excavator, build a dry-stone wall from boulders found on site (Science Robotics, 2023). Boulders do not flinch. Units that reach real sites tend to ship with a teleop fallback, and the business case turns on operator minutes per machine hour. The operators are rarely credited.

Teleoperation buys capability today at the price of one trained human per machine. From my vantage in the late 2070s, the site robots that failed were not the weak ones; they were the ones nobody on the crew could repair or reset.

←TODAY: In September 2026 a remote-controlled humanoid kicks a padded influencer across a cage in a 44-second clip; nobody names the pilot.
→3012: Zurich’s site robots read the building model before entering a room, write back what they changed, and carry a manual reset an apprentice can reach.
Fulcrum: The kick is the demo; the hand on the controller is the product, and the model the next robot reads is the one we draw.

An octagon is empty. A building is not.

A building is corridors, door leaves, thresholds and a lift that is out of service on Tuesdays. Robot vacuums already map flats with lidar from scratch, bumping into the Türrahmen an architect drew years earlier. A robot on a hospital ward or a Bauleitung site should read the model instead and, across the building’s life, write back what changed: a twin in the Digital Twin Consortium’s sense, synchronised at a specified frequency and fidelity.

The reading stack is open and named: ROS 2 under the Open Source Robotics Foundation, Nav2 led by Steve Macenski, MoveIt 2 maintained by PickNik Robotics. Their labour is invisible because it works.

Atelier: The first robots in your buildings will not fight anyone; they will clean, inspect and carry, navigating by whatever model you handed over, or by nothing. For an office reorganising around AI, that makes the IFC export instructions for machines as well as drawings for people. Monday move: add one BEP line requiring every IfcSpace exported with a name and every IfcDoor with its width populated.

Hack: Pull every door a robot chassis cannot pass straight out of the IFC with IfcOpenShell, the open library Thomas Krijnen started and Dion Moult and many contributors keep running. The unit scale turns any Archicad or Bonsai export into metres, so the 0.80 m clear width SIA 500 asks of barrier-free doors holds on every file. OverallWidth measures the whole opening, not the clear passage, so a door just above the line can still pinch a chassis. Run it on a live project: one pass gives a robot’s first route plan and a barrier-free check.

import ifcopenshell, ifcopenshell.util.element as E, ifcopenshell.util.unit as U
m = ifcopenshell.open("site.ifc"); k = U.calculate_unit_scale(m)
for d in m.by_type("IfcDoor"):
    if d.OverallWidth and d.OverallWidth * k < 0.80:
        print(getattr(E.get_container(d), "Name", "?"), d.Name, round(d.OverallWidth * k, 2))

Next time a robot video hits the office chat, ask what a site unit asks first: who holds the controller, and what happens when they let go? Then run the door check on your latest IFC.

Source: futurism.com