Collaborative Robot Safety: What Replaces the Cage

For forty years, industrial robot safety was a fence. Put the machine inside, put the people outside, interlock the gate. It worked, and the reason it worked is that it made no assumptions about the robot’s software. Collaborative robot safety is what replaced it, and it’s a set of four named engineering methods rather than a philosophy. Three of them depend on sensors and controllers behaving correctly. One of them doesn’t. Knowing which is which is the single most useful thing a parent can understand about why a robot in a warehouse is safe and why nobody has fully solved the same problem for a walking humanoid.

Key Takeaways

• OSHA’s Technical Manual names four safeguarding methods for collaborative operation: safety-rated monitored stop, hand-guided control, speed and separation monitoring, and power and force limiting.
• Only power and force limiting is a physical guarantee. It keeps contact forces inside safe biomechanical limits by design, so the robot cannot hurt you much even if its software is wrong.
• The governing standards OSHA cites include ANSI/RIA R15.06, ISO 10218-1 and ISO 10218-2, ISO/TS 15066, and ANSI/A3 R15.08 for mobile robots.
• The hazards haven’t changed, only the controls. NIOSH documented a fatality back in 1984 in which a worker “entered the working range of the robot to do maintenance and was pinned between the back of the industrial robot and a steel safety pole.”
• Walking robots break the oldest safety trick in the book. Cutting power to a bolted-down arm stops it. Cutting power to a balancing biped makes it fall. Agility’s Digit 5 answers that by setting its load down and sitting.

Why the cage worked, and why it had to go

A safety cage works because it is a physical barrier whose effectiveness doesn’t depend on any software being correct. That’s the property to hold onto, because everything that followed is a negotiation with it.

What a cage can’t do is let a person and a robot share a task. Plenty of manufacturing work is genuinely collaborative: a human orients a part while the robot drives a screw, or a human inspects while the robot holds. Fencing that off means either two separate stations with handoffs, or no automation at all. The economics pushed hard on this, and the standards followed.

OSHA’s definition of what’s actually being made safe is broader than people expect. The industrial robot system includes “not only the industrial robot but also the end-effector attached to the robot manipulator; computers, processors, and programs (i.e., the control system); power sources; sensors; and, sequencing or monitoring communication interfaces (i.e., input/output devices).” The arm is one component of the hazard.

And the hazard list is unglamorous: impact and collision, crushing and trapping, being struck by projectiles when an end-effector or workpiece fails, electrical, hydraulic and pneumatic energy, slips, trips and falls, and environmental exposures like chemicals, heat, noise and radiation. That NIOSH case from December 1984, a maintenance worker pinned between a robot’s back and a steel pole, is the canonical shape of the problem. The person wasn’t in front of the robot. They were behind it, where nobody was watching.

The four methods, and which ones are promises versus guarantees

Safety-rated monitored stop. The robot detects a person entering a defined space and stops, while keeping power to the actuators. OSHA describes this as a “Category 2 Stop,” which is the important detail: the motors stay energized and actively hold position. Cutting power would let gravity move the arm. The robot resumes when the person leaves.

Hand-guided control. A worker physically moves the robot using a hold-to-run device, release the control and motion ends. The human is the controller, which is why this is permitted at close range.

Speed and separation monitoring. Sensors track the distance between person and robot, and the robot slows as the gap closes and stops before contact becomes possible. This is the method that enables genuinely fluid shared workspaces, and it’s the most software-dependent of the four.

Power and force limiting. The robot is designed so that any contact stays within safe biomechanical limits, through the design itself or through safety-rated functions. ISO/TS 15066 is the document that specifies those limits by body region.

Now the part worth screenshotting. If a controller is compromised or simply buggy, safety-rated monitored stop can fail to trigger, hand-guided control can misread the dead-man switch, and speed and separation monitoring can misjudge a distance. A genuinely power-and-force-limited robot still can’t hurt you much, because the actuators physically cannot produce enough force. One of these four is a property of the hardware. The other three are properties of a system working as intended. We develop that distinction further in robot cybersecurity and the problem nobody budgeted for.

What walking robots do to all of this

Every method above assumes the robot is either bolted down or wheeled. Legs break the assumptions.

The deepest problem is the emergency stop itself. For a fixed arm, removing power is the safest possible action. For a dynamically balancing biped, removing power means a hundred-plus kilograms falls over, which is worse than whatever prompted the stop. The oldest and most trusted safety mechanism in industrial automation has the wrong sign.

IEEE Spectrum reported in September 2025 that new ISO standards for dynamically balancing legged robots were still under development, with Boston Dynamics’ Matt Powers, associate director of autonomy R&D, describing the company’s intent to “start with relatively low-risk deployments, and expand as we build confidence.” That’s the honest state of play: the robots exist, the standard doesn’t yet.

Agility’s answer with Digit 5, covered September 15, 2026, is the most interesting engineering response so far. The robot autonomously avoids, stops, or assumes a seated position. When a person approaches, it puts its load down and sits. Evan Ackerman’s assessment was carefully bounded: the approach “works, which is more than can be said for any other commercial humanoid.” A seated 129-kilogram robot is furniture. A falling one is an incident report.

How to Teach Your Kid About Collaborative Robot Safety

These are genuinely good physical demonstrations, and they stick.

Ages 5–8: Tape the restricted space

Pick a spinning or swinging object: a desk fan on a table, a pendulum on a string. Tape a line on the floor at the edge of where it can reach. OSHA calls that volume the restricted space: where the machine operates and a person may encounter the hazard. Have your kid walk the line and notice that safety is about geometry first. Then move the fan and ask: where does the tape go now?

Ages 9–12: Measure a safe force

Get a kitchen scale. Have your kid press on it with one finger until it’s uncomfortable, and write the number. Then try the back of the hand, then the forearm. The numbers differ a lot, and so do the limits in ISO/TS 15066, which specifies allowable forces by body region. That’s the entire concept of power and force limiting, measured on a kitchen counter in ten minutes.

Ages 13+: Design the stop

Give them a scenario: a wheeled robot carrying a hot cup of coffee down a hallway, and a person steps out of a doorway. Have them design the response using each of the four methods, then identify what each one requires to work: which sensor, which timing, which assumption. Then the hard question: which of their four designs still works if the robot’s main computer crashes? Only one should survive, and figuring out which is the lesson.

The question to ask: “If a robot’s computer stops working, which kind of safety still protects you?”

The four collaborative safety methods, compared

Method	How it protects	Depends on software?	Survives a controller failure?	Typical use
Safety-rated monitored stop	Robot halts but stays powered when a person enters the space	Yes, detection and stop logic	No	Shared station with alternating access
Hand-guided control	Human moves the robot via a hold-to-run device	Partly, the enabling device is safety-rated	Partly	Teaching positions, assisted lifting
Speed and separation monitoring	Robot slows and stops as the gap closes	Heavily, ranging sensors plus control	No	Fluid shared workspaces
Power and force limiting	Contact forces stay within safe biomechanical limits	Can be purely mechanical	Yes, when designed into the hardware	Small cobots working directly with people
The old steel cage	A physical barrier with an interlocked gate	No	Yes	High-speed, high-payload cells

Notice that the cage is still the sturdiest row in the table. It’s not obsolete: it’s still the right answer for a fast, heavy robot doing a repetitive task, and that describes most of the 5 million industrial robots the International Federation of Robotics counted in operation as of 2025. Collaboration is the exception, not the replacement.

What this means for a family and a school

Weight and torque are the specs you can actually verify

You cannot audit a vendor’s safety software. You can read a mass and a motor rating off a spec sheet. A small, low-torque educational robot is in a different risk class than a large one regardless of its firmware, and that’s the parent-accessible version of power and force limiting.

Ask which standard a school robot was built to

ANSI/RIA R15.06, ISO 10218-1 and -2, ISO/TS 15066 for collaborative operation, and ANSI/A3 R15.08 for mobile robots are the names to listen for. A vendor selling into schools who can’t name a standard has told you the robot was designed to no standard in particular.

Teach that behind the robot is as dangerous as in front

The 1984 NIOSH case is a maintenance worker pinned at the back of a machine. Robots have working envelopes, not front sides, and a kid who learns to think about the whole swept volume has internalized the most common cause of real robot injuries.

Treat maintenance as the risky moment

Almost every documented industrial robot incident involves someone inside the envelope doing maintenance, programming or clearing a jam, not normal operation. If a kid is ever near a working robot in a shop class or a competition pit, the rule is simple: the dangerous moment is when it’s stopped and someone’s reaching in. Our piece on collaborative robot programming as a career covers who does this work professionally.

What not to do

Don’t tell a kid a collaborative robot is “safe.” Tell them it’s safe under conditions, and have them name the conditions. A cobot with a sharp tool mounted on it is not safe regardless of its force limits, because ISO/TS 15066’s biomechanical limits assume blunt contact. The end effector changes the risk completely, which is exactly why OSHA defines the robot system to include it.

What to Watch For Over the Next 3 Months

• Week 4: Watch for publication or balloting news on an ISO standard covering dynamically balancing legged robots. As of the September 2025 reporting it was still in development, and its arrival would be the gating event for humanoids working near people.
• Month 2 red flags: Any humanoid described as “safe” with no mention of a safeguarding method, a force limit or a standard. Also any demo where a person stands close to a walking robot with no visible barrier and no stated protocol: that’s a choreographed shot, not a safety claim.
• Month 3 self-check: Ask your kid which safety method they’d want if they had to work next to a robot every day. If they pick power and force limiting and can say why, they’ve understood the difference between a guarantee and a promise.

Frequently Asked Questions

Are collaborative robots actually safe to work beside?

Under defined conditions, yes, and the conditions do the work. The robot has to be assessed for its specific application, including its end effector, its speed and the task, under the relevant standard. The same cobot can be safe holding a plastic part and unsafe holding a blade.

Why can’t you just use an emergency stop on everything?

Because an emergency stop removes power, and for a dynamically balancing robot that causes a fall. That’s why OSHA’s safety-rated monitored stop is a Category 2 stop that keeps actuators energized, and why legged robots need an entirely different concept, like Digit 5 sitting down with its load placed.

Does a robot vacuum or lawn mower at home follow these standards?

Consumer products fall under different regulatory regimes than industrial robot standards, and the OSHA framework in this article governs workplaces. The underlying physics still applies: a low-mass, low-torque machine has a low harm ceiling, which is why household robots are built that way.

How dangerous are industrial robots in practice?

Serious incidents are rare relative to the installed base, and they cluster heavily around maintenance and programming rather than normal operation. The hazard pattern OSHA documents (impact, crushing, trapping, projectiles from end-effector failure) reflects people being inside the envelope when they shouldn’t be.

Will the cage ever come back?

It never left. For a fast, heavy robot doing repetitive work, a fence remains the cheapest and most reliable safeguard, and it’s still standard across most of the millions of industrial robots in service. Collaborative methods exist for the subset of tasks that genuinely need a human in the loop.

What should a kid learn if they want to work on this?

Functional safety is a real specialization with real demand: risk assessment, safety-rated sensors and controllers, and standards literacy. It sits between mechanical engineering, controls and regulatory work, and almost no one chooses it early. Our comparison of robotics engineer versus robotics technician covers the surrounding paths.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Occupational Safety and Health Administration. “Industrial Robot Systems and Industrial Robot System Safety.” OSHA Technical Manual, Section 4, Chapter 4. https://www.osha.gov/otm/section-4-safety-hazards/chapter-4
2. Occupational Safety and Health Administration. “Robotics — Hazard Recognition.” https://www.osha.gov/robotics/hazards
3. National Institute for Occupational Safety and Health. (1984, December). Preventing the Injury of Workers by Robots. DHHS (NIOSH) Publication No. 85-103 — as cited and quoted on OSHA’s robotics hazard recognition page. https://www.osha.gov/robotics/hazards
4. Ackerman, E. (2026, September 15). “Digit 5 May Be the First Humanoid Robot Worker That’s Truly Safe.” IEEE Spectrum. https://spectrum.ieee.org/humanoid-robot-safety
5. Ackerman, E. (2025, September 11). “Reality Is Ruining the Humanoid Robot Hype.” IEEE Spectrum. https://spectrum.ieee.org/humanoid-robot-scaling
6. International Federation of Robotics. (2026, September 24). “Five Million Robots now Operate in Factories Globally.” https://ifr.org/ifr-press-releases/news/five-million-robots-now-operate-in-factories-globally
7. VicOne (sponsored). (2026, September 16). “Rethinking Robot Safety in the Age of AI.” IEEE Spectrum. https://spectrum.ieee.org/physical-ai-robot-cybersecurity-vicone