Collaborative Robot Risk: Assess the Task, Not the Arm
Direct answer: Power and force limiting addresses impact between the arm and a person. It does not address what the task brings: pinch points against a moving conveyor, reaction torque at a tightening spindle, multi-pin tooling, hot parts, and process media such as high-pressure air or mist. ISO 12100:2010 builds the assessment from the application’s tasks and hazards, ISO 10218-2:2025 covers the integrated cell, and ISO/TS 15066:2016 supplements them. A light curtain is a protective measure, not the assessment.
Who this is for: Written for engineers and safety officers specifying collaborative robot applications where people can reach the working area.
Scope: This EVST guide covers why a power- and force-limited arm does not by itself make an application safe, and what a task-based assessment has to cover instead. It does not state force limits, protective distances or any figure for a specific installation; those come from the assessment for that application.
Force limiting solves one hazard
A power- and force-limited robot restricts the forces and pressures that can arise in contact between the arm and a person. That is a real and useful property, and it is the property most often used to justify removing a fence.
It is also a narrow property. It says nothing about what the arm is carrying, what the part is doing, how hot it is, or what the process releases into the space.
The three cells in the reference footage make the point without argument: the same class of arm appears in all three, and the dominant hazard is different every time.
What a collaborative robot task based risk assessment covers
The assessment starts from tasks, not from equipment. For each task performed at the cell, including setup, clearing a fault and cleaning, the hazards are identified and then evaluated.
According to ISO 12100:2010, hazard identification is carried out for the tasks actually performed at the machine over its whole life cycle, which is why an assessment built only around normal running is incomplete.
According to ISO 10218-2:2025, it is the integrated cell that is assessed, so the tool, the part, the conveyor and the surrounding equipment are inside the boundary rather than outside it.
According to ISO/TS 15066:2016, collaborative operation is addressed as a supplement to ISO 10218 rather than as an alternative to it, so a force-limited arm does not remove the need for the assessment.
EVST starts a review from the task list, because the arm’s rating is the one part of the picture that is already known.
The tool changes the hazard
In the first cell the end effector is a row of pins placing fasteners onto a part that is moving on a line. The arm may be limited, but the geometry between tool, part and conveyor creates pinch points that limiting does not address.
In the second cell the same class of arm carries a tightening spindle. Now the dominant hazard is reaction torque and what happens if the tool grabs, which is a different mechanism with a different protective answer.
The table below sets out that relationship for the common cases.
| Task at the cell | Hazard beyond arm impact | What to verify |
|---|---|---|
| Placing parts onto a moving line | Pinch between tool, part and conveyor | Clearances at every point of the stroke, at line speed |
| Screw or bolt tightening | Reaction torque and tool behaviour on a grab | Torque path, tool restraint and where the reaction is taken |
| Multi-pin, bladed or hot tooling | Sharp geometry, trapping and burns | Tool geometry, approach speed and where hands can be |
| Process using air, mist, coolant or heat | The process media, not the arm | Whether the process alone justifies enclosure |
A guard is a measure, not an assessment
In the second cell a light curtain is fitted at the front of the station. Protective equipment being present is evidence that someone identified a hazard; it is not evidence that the assessment is complete or that the residual risk is acceptable.
The order matters. Hazards are identified, risk is estimated, and measures are then selected and verified against that estimate. Reading it backwards, from the measure to the conclusion, is how a cell ends up guarded against the hazard someone thought of and open to the one they did not.
According to ISO 13857:2019, safety distances are established with respect to the hazard zone being protected, so a device only performs as intended when it is positioned against the specific hazard it was chosen for.
When enclosure is about the process
The third cell in the footage is fully enclosed, with the robot working behind a window. The enclosure is not there because the arm is dangerous; it is there because the process uses high-pressure air and produces water mist, and neither belongs in an occupied space.
That distinction is worth keeping explicit, because it is easy to read an enclosed cell as evidence that collaborative operation failed. It is more often evidence that the process was assessed on its own terms.
The practical consequence is that the choice between an open and an enclosed cell is not a robot decision. It follows from the hazards of the task, which is the same conclusion the first two cells reach from the opposite direction.
What belongs in the file
The assessment itself, the task list it was built from, the force and pressure values used where power and force limiting is relied on, and the verification of the protective measures all belong in documents rather than in a description of the cell.
None of that is visible in footage, and no reading of the video should be taken as evidence about it. What footage can show is the shape of the cell, the tools in use, whether people can reach the working area and what the part is doing.
In practice an EVST application review starts from the same project inputs each time: the cell layout, the tasks performed there, the part temperature and how people move past. Selection and safeguarding follow from that evidence rather than from a datasheet.
Frequently asked questions
Does a collaborative robot remove the need for a risk assessment?
No. ISO/TS 15066:2016 supplements ISO 10218 for collaborative operation rather than replacing it, and ISO 12100:2010 still requires the assessment to be built from the tasks performed at the application.
If a light curtain is fitted, is the cell assessed?
Not necessarily. A protective device shows that a hazard was identified, not that all of them were, nor that the device is positioned correctly for the hazard it is meant to cover.
Why would a collaborative arm be put inside an enclosure?
Usually because the process, rather than the arm, is the hazard. High-pressure air, mist, coolant, heat or debris can each justify enclosure independently of how the robot is rated.
What should we send to get a useful answer?
The cell layout with the surrounding equipment, the list of tasks performed there including setup and fault clearing, the part temperature, and how people walk past or reach in.
Project inputs for an application review
To have an application reviewed against its own tasks rather than a general rating, send:
- cell layout including conveyors and adjacent equipment
- the task list, including setup, cleaning and fault clearing
- end effector description and part temperature
- how people approach, pass or reach into the working area
Send the cell layout, the task list and the part temperature and the application can be worked through against them. Related reading: collaborative and industrial robots, robot workstation integration.