What Are the Cobot Safety Standards?

The main cobot safety standards are ISO 10218, the international safety standard for industrial robots, and ISO/TS 15066, a technical specification for collaborative operation. ISO 10218 comes in two parts: Part 1 covers the robot itself and is written for robot manufacturers, while Part 2 covers the robot system and its integration into a cell or application, and is written for integrators and users.

ISO/TS 15066 adds guidance specific to robots that work alongside people, including limits on the force and pressure a robot may apply if it touches different parts of the body. Both parts of ISO 10218 were revised in 2025, and the new Part 2 brings much of the collaborative guidance from ISO/TS 15066 into the main standard.

One principle runs through all of them: a robot is not safe on its own. Safety belongs to the application, which includes the tool, the part being handled, the layout and the people nearby. Every application needs its own risk assessment.

In the US, ANSI/RIA R15.06 adopts ISO 10218, and in Canada CSA Z434 does the same. In Europe the standards are harmonised under the Machinery Directive, which the new Machinery Regulation replaces from January 2027, so following them gives a presumption of conformity.

The standards describe four ways a robot can collaborate safely. An application may use one or combine several.

For power and force limiting, ISO/TS 15066 gives a body model with permissible force and pressure values for different body regions, from the hand and arm to the chest and face. Sensitive areas such as the face, skull and neck have much lower limits, and many applications are arranged so that contact with them cannot happen at all.

The specification separates two kinds of contact. Transient contact is a brief impact where the person can move away. Quasi-static contact is clamping or crushing, where a body part is trapped between the robot and something else, which has lower permitted values. Integrators measure the real forces and pressures with a test device during validation, rather than relying on settings alone.

Whatever the method, the application starts with a risk assessment under ISO 12100. It identifies each hazard, from the sharp edge on a gripper to the part being carried, and decides how to reduce the risk. The safety functions that result, such as speed limits, zone monitoring and stops, are designed to a required performance level under ISO 13849-1, typically PL d for robot safety functions.

The cobot label on the robot does not remove this step. A cobot carrying a knife or a heavy casting can be just as dangerous as a conventional robot, and may need guarding or scanners like one. The assessment, the measured contact forces and the validation of each safety function are recorded in the technical file, which is reviewed whenever the task, tool or layout changes.

Vision systems increasingly support speed and separation monitoring. Cameras and AI models running on an edge computer can track where people are and how they move, using techniques such as pose estimation, so the robot slows earlier and stops less often.

Unless the vision system is itself certified as a safety-rated sensor, it works alongside certified safety devices such as laser scanners and light curtains, not instead of them. The safety-rated layer keeps people safe; the AI layer keeps the line productive. Our human-robot collaboration case study shows this approach on NVIDIA Jetson, and our explainer on the robot controller covers where the robot’s own safety logic runs.

For vision-assisted robot cells, we supply rugged Jetson computers to run perception alongside the certified safety layer, integrated and validated as standard before they ship.