The humanoid robot safety question nobody's asking

A Figure 02 robot weighs about 130 pounds. A Tesla Optimus weighs about 125 pounds. A Boston Dynamics Atlas (the new electric one) weighs about 190 pounds.

These are machines that weigh as much as an adult human, move through the same spaces as humans, and carry objects that could injure someone if dropped. They have arms that swing. Legs that step. Hands that grip with force measured in pounds per square inch.

What are the safety standards?

Practically none.

The gap

We have ISO standards for industrial robot arms that operate behind cages. We have OSHA regulations for factory machinery. We have UL certifications for every electrical device in your home.

For humanoid robots that walk freely among people, we have… guidelines. Recommendations. Research papers. Nothing binding. Nothing that a company must comply with before putting a 170-pound autonomous machine next to a human worker.

This isn’t a hypothetical. Figure is deployed in a BMW factory. Agility’s Digit works in Amazon warehouses. Tesla plans consumer sales in 2027. These machines are entering workplaces and will soon enter homes.

The scenarios I think about

A robot carrying a 30-pound box trips on uneven ground and the box strikes a worker. Who’s liable? What force limits should the robot have? What’s the acceptable failure rate?

A robot misreads a human gesture. Someone waves their hand to say “stop” and the robot interprets it as “come closer.” At walking speed, 130 pounds of metal and actuators arrives before the person can move.

A software update introduces a regression in the balance algorithm. The robot falls. It falls onto someone. The impact force of a 130-pound machine falling from standing height is significant.

These aren’t edge cases. They’re Tuesday. They’re the kinds of things that happen in environments where machines and humans share space. The automotive industry has a century of safety engineering to address these scenarios. The robotics industry is building safety standards in parallel with deployment.

That’s backwards. Safety standards should lead deployment, not follow it.

What needs to happen

Force limits. Every humanoid robot should have hard limits on the force its limbs can exert. Not software limits (software can bug). Hardware limits. Physical stops that prevent an arm from swinging with more force than a defined threshold.

Collision detection with immediate shutdown. If a robot contacts a person unexpectedly, it should stop moving within milliseconds. Not slow down. Stop.

Mandatory incident reporting. Every robot-human contact event should be reported to a centralized database. The way aviation reports near-misses. The data builds the safety record.

Third-party certification. Before a humanoid robot can operate alongside humans, an independent body should verify its safety systems. The way cars get crash-tested. The way planes get certified.

I’m not anti-robot. I’ve been writing about the promise of humanoid robots for years. But promise without safety is recklessness. And right now, the safety infrastructure is far behind the capability.

The time to build it is now. Not after the first serious injury. Now.

Related thinking: