English
While the world argues about humanoid robots and self-driving cars, a quieter class of machine has slipped into some of the most dangerous workplaces on Earth: the radioactive room, the gas-filled mine, the collapsed building. They walk on four legs, they don't need to breathe, and they're already doing the jobs we should never have sent humans to do.
There are rooms in this world that will kill a person for walking into them. A chamber inside a damaged nuclear plant, thick with invisible radiation. A mine shaft filling silently with gas that stops a heart before the brain registers danger. The unstable guts of a collapsed building, one wrong step from a second collapse. For all of human history, checking these places meant risking (or spending) a human life.
Increasingly, it doesn’t. Something else goes in first. It walks on four legs, it carries cameras and sensors instead of lungs, and it does not care that the room is trying to kill it.
The robot that looks like a dog and works like a canary
The best-known of these machines is Spot, made by Boston Dynamics, the yellow, dog-sized quadruped you may have seen in unsettling viral videos, trotting up stairs and recovering from a kick. The videos made it look like a novelty. It isn’t. Spot, and rivals like ANYmal from the Swiss company ANYbotics, have quietly become working industrial equipment.
Their day job is inspection. Oil and gas facilities, power plants, factories, and substations are full of gauges to read, valves to check, pipes to scan for heat or leaks, and machinery that hums along fine until, unwatched, it doesn’t. Traditionally a human walks a route with a clipboard and a thermal camera, day after day, including through areas that are hot, loud, cramped, or quietly hazardous. Now a legged robot walks the route instead: the same path, at any hour, uncomplaining, uploading readings and flagging anything wrong. It’s a canary in a coal mine that can also read the dials and file the report.
One detail gets lost in the marketing, and it matters: not every one of these machines is cleared for every hazard. Sniffing for gas in a room that could ignite is a specialist job, and ANYbotics builds a specific model for it, ANYmal X, certified under the ATEX and IECEx standards for Zone 1 explosive atmospheres and fitted with gas detection. A standard Spot or a standard ANYmal does not carry that rating. Send the wrong model into the wrong room and the robot is the spark.
Why legs, of all things
It’s a fair question. We’ve had wheeled robots for ages, so why grow them legs, which are mechanically complicated and famously hard to balance?
Because the dangerous world was not built flat. It was built for people, which means it’s full of stairs, ladders, curbs, thresholds, gratings, debris, and narrow gaps: terrain that stops a wheeled robot dead. A machine that can only roll is confined to smooth floors. A machine that can step, climb, clamber over a pipe, and pick its way across rubble can go where the danger actually is.
That’s the whole case for the legged form. Legs are a pain to engineer, but they buy access to the exact environments (industrial, broken, improvised) where sending a human is worst. The complexity isn’t showing off. It’s the price of getting into the room.
When the room is already a disaster
The most vivid use is disaster response, where the environment is more than hazardous: it’s actively hostile and unknown.
After an industrial accident, an earthquake, or a structural collapse, the first and most agonising question is: can anyone go in, and is anyone alive? Sending human rescuers means gambling more lives on unstable ground, in air that may be toxic, in spaces that may collapse again. A four-legged robot can be sent ahead to map the interior, look for survivors, test the air if it is the explosion-rated model carrying a gas sensor, and stream back a picture of what’s really in there, turning a blind, deadly gamble into an informed decision.
Radiation is the case people get wrong. It’s tempting to say a robot can walk into a hot zone, soak up a dose that would kill a person, and carry on. It can’t. Electronics are more fragile than flesh in this one respect: radiation quietly wrecks wiring, chips, and sensors, and lab testing on reconnaissance robots found the cameras and lidar give out first, at around 120 gray. Fukushima Daiichi is the hard lesson. Machines sent into the reactor buildings failed within hours as the radiation ate their insides, and one was built to survive a total of 73 sieverts before it met a measured field of 530 sieverts an hour.
So the real engineering fight is radiation hardening: shielding the electronics, moving the vulnerable parts away from the hot side, and building the machine to finish one useful run rather than live there. And even when the robot loses, the argument holds. A machine that dies inside the reactor building has mapped it, filmed it, and taken readings nobody could otherwise get. It leaves behind an invoice, not a widow.
This is robotics doing the thing robotics was always supposed to do, and too rarely gets credit for: taking the jobs that are dull, dirty, and above all dangerous, and getting flesh-and-blood humans out of harm’s way.
The unglamorous frontier
None of this makes headlines the way a humanoid or a robotaxi does. There’s no drama in a machine successfully reading a gauge, or in a mine inspection where nothing went wrong precisely because a robot, not a person, found the problem first. Success here is invisible: an accident that didn’t happen, a rescuer who didn’t have to risk it, a dose of radiation that landed on a machine instead of a body.
But that’s arguably the most humane corner of the whole robotics boom. Not the robots trying to look like us, or drive like us, or replace us, but the ones walking, right now, into the rooms that would kill us, so that no one else has to.
Sources & further reading
Researched and written with the help of AI tools and edited for accuracy. Provided for general information and discussion only, not professional advice. See our editorial standards and disclaimer. Spotted an error? Tell us.
Enjoyed this? Get the next one.
One good read at a time, straight to your inbox. No spam, unsubscribe anytime.
