Why does your robot safety case still stall on the factory floor?
You bought a collaborative robot, and the datasheet says it is safe. Yet your floor team still fences it off, and nobody can say what would change their mind.
The standard sets the floor, not the case
ISO 10218 is the international standard for industrial robot safety, developed by ISO's robotics committee and published in two parts, one for robots and one for robot cells and applications. Per Wikipedia's summary, the 2025 editions replaced the 2011 editions in February 2025. A certificate shows you met the minimum. It does not show your specific cell, with your parts and your people, behaves predictably when something fails.
- ISO 10218 Part 1 covers robots; Part 2 covers applications and cells
- The 2025 editions cancel and replace the 2011 editions
Cobots moved the line in 2016
Wikipedia notes ISO/TS 15066 was published in 2016 to address safety requirements for cobots, robots meant for direct, close-range human interaction. Before that, the model was simple: isolate the robot or keep people away. Cobots invert it. Safety now depends on lightweight materials, rounded edges, and limits on speed and force, or on sensors and software that ensure safe behavior.
Functional safety is end-to-end
Per Wikipedia's functional safety article, functional safety means automatic protection operates correctly in response to inputs, or fails in a predictable, fail-safe manner. It is intrinsically end-to-end: a component only counts as safe as part of the entire protection function. That is why a certified arm plus a generic sensor setup still leaves gaps. The safety case must cover the whole cell, including how it fails.
What this does not solve
These sources describe standards and definitions, not outcomes. Nothing here tells you whether a given safety case convinced a real factory, what it cost, or how auditors judged it. If your problem is stakeholder trust, standards give you vocabulary, not evidence. You still need incident data and cell-level validation of your own.
Which failure modes in your cell have you actually tested end-to-end, versus assumed from component certifications?
Who on your floor would need to see what evidence before they stopped fencing off your cobot?
List every safety function in one robot cell and mark which ones you have tested as a whole system, not just as parts.
Sources
Recommended for you
- Robotics
What breaks when a robot meets the real world?
Your AI passes every test but trips over a doormat. Perception, not reasoning, is where robots fail, and budgets miss it.
- Robotics
Why does robot training data still cost so much?
Robots learn by trial and error, and every trial happens in the physical world. Even digital twins, the promised shortcut, need real machines feeding them data.
- Robotics
Why do humanoid robots still have no proven business model?
Billions in valuation, but no revenue story: the humanoid robot race runs on promises of generality, not paying customers.
- Robotics
Why does robot payback math still mislead buyers?
Vendors now quote payback periods from months to decades. The spread is the story: the number depends on what you count, not on the robot.
Enjoyed this article?
Subscribe for new articles. No spam. Unsubscribe anytime.