Stopping, limiting, and separating: the core toolkit
Underneath the standards, a fairly small set of recurring engineering mechanisms actually does the work of making robots safer, and it's worth naming them concretely rather than treating 'safety' as an abstract compliance checkbox. Emergency stop systems, both hardware, a physical button that cuts power or commands an immediate halt, and software, a monitored stop condition triggered by sensor input, exist as the most basic layer, a way to halt the robot fast when something has already gone wrong. Safety-rated speed and force limiting caps how fast a robot can move or how much force it can exert in specific zones or modes, directly reducing the severity of any contact that does occur, which is the mechanism underlying much of ISO/TS 15066's power and force limited operation. Protective separation, light curtains that detect a limb breaking a beam, safety-rated fencing, and presence-sensing floor mats or laser scanners, prevents a person from entering a hazardous zone at all while the robot is operating in a mode that isn't safe for human presence.
These mechanisms are typically layered rather than used alone, because a robot safety system generally shouldn't depend on a single point of protection working perfectly every time. A well-designed cell might combine speed limiting near a collaborative zone, presence sensing that triggers a slowdown before full stop is needed, and a hardware e-stop as the last-resort backstop, so that if one layer is defeated or fails, another is still in place. That layering, not any single mechanism, is closer to how real safety-critical engineering actually thinks about risk.