Collaborative Robots
Cobot Risk Assessment: The Steps Auditors Check
A working sequence for a collaborative robot risk assessment, from limits of the machinery to validated measurement, and the six omissions that come back as audit findings.

A collaborative risk assessment is the same ISO 12100 process every machine gets, with one extra loop: after choosing protective measures you must classify every reachable contact, derive a permissible speed, and then prove it by measurement. Auditors do not challenge the method, they challenge the evidence, and the six omissions listed below account for most findings.
The sequence
- Determine the limits of the machinery. Spatial limits including full reach with tooling, time limits including maintenance and cleaning, and use limits including foreseeable misuse. This is where most assessments are already too narrow: the swept volume with a 400 mm gripper is not the swept volume on the datasheet.
- Identify hazards across the whole life cycle. Production is the easy part. Teaching, tool change, part jam clearing, cleaning, maintenance and decommissioning each have their own exposure, and jam clearing is where people actually get hurt.
- Estimate and evaluate risk. Severity, frequency of exposure, possibility of avoidance. Document the reasoning, because the conclusion is worthless without it.
- Apply the hierarchy. Inherently safe design first, then technical protective measures, then information for use. Reducing mass, rounding geometry and removing pinch points belong in step one of the hierarchy, not in a warning sign.
- Choose the collaborative operation type per task segment. Most cycles use two, for example speed and separation monitoring during transfer and contact-permitted operation only at handover.
- Classify every reachable contact. For each: body region, transient or quasi-static, contact area, and the resulting limit values.
- Specify and verify safety functions. Each function gets a required performance level, an architecture, and validation that it behaves as specified, including on fault injection.
- Measure. Force and pressure at every reachable contact point with a calibrated instrument and a spring element matched to the body region. Record instrument, date, configuration, speed and payload.
The contact table is the heart of the file
If an auditor reads one page, it is this one. A workable format has one row per reachable contact point and no gaps.
| ID | Location | Body region | Class | Limit N | Limit N/cm² | Measured |
|---|---|---|---|---|---|---|
| C1 | Gripper front face, approach to nest | Hand and fingers | Transient | 280 | 520 | 191 N |
| C2 | Jaw closing zone | Hand and fingers | Quasi-static | 140 | 260 | 118 N |
| C3 | Elbow sweep over bench edge | Lower arm and wrist | Quasi-static | 160 | 180 | 143 N |
| C4 | Upper arm sweep at operator side | Chest | Transient | 280 | 240 | 205 N |
| C5 | Gap between arm and machine frame | Upper arm and elbow | Quasi-static | 150 | 190 | designed out |
Row C5 is the model answer. A clamping gap that would have needed a measurement was removed by moving the machine 120 mm, and the row records that decision. Designing a scenario out is always stronger evidence than measuring it and passing narrowly.
The six omissions auditors find
- The tool is not in the reach study. Reach is assessed at the flange, so a long gripper or a part sticking out beyond the jaws extends the swept volume by 200 mm to 500 mm that nobody assessed.
- Jam clearing is missing. Production is assessed, intervention is not, yet intervention is when a person is closest to the machine and most distracted.
- Quasi-static scenarios classified as transient. This doubles the apparent allowance and is the single most consequential error in the file, because the real limits are half what was assumed.
- Measurement predates the final program. Speeds were raised during commissioning after the measurement was taken. Any change to speed, payload or tooling invalidates the report.
- No performance level derivation. Safety functions are listed but the required level is asserted rather than derived, and the fault behaviour was never tested.
- Head reachability unaddressed. A seated operator, a stool, or a maintenance step changes the geometry, and contact with the head is never permissible.
When it has to be repeated
The assessment covers a configuration, not a machine. Repeat it, or at minimum re-measure, whenever any of these change: programmed speed, payload or gripper, cell layout or fixture position, the part family, the operating type used in a segment, or the software version that controls a safety function. A new part variant that is 300 g heavier is a change to effective mass, and effective mass sets the permissible speed.
For machinery placed on the European market, the applicable legal frame is fixed by the date of placing on the market: Directive 2006/42/EC until 19 January 2027, Regulation (EU) 2023/1230 from 20 January 2027, with no overlap.
Frequently asked questions
Who has to carry out the risk assessment?
Whoever places the complete cell into service, usually the integrator together with the operating company. The robot manufacturer assesses the robot as partly completed machinery, which does not cover your gripper, fixture and layout.
Is a vendor safety certificate enough?
No. It applies to the arm in isolation. The moment the arm carries a tool and faces a workpiece, a new machine exists and it needs its own assessment, its own validation and its own declaration of conformity.
How long does a collaborative risk assessment take?
For a single-station cell, typically two to five days of engineering including the contact table, plus about a day of measurement for a dozen reachable contact points. Cells with high part variety take considerably longer.
Do I have to measure, or can I rely on the vendor's data?
Vendor data describes the arm, not your assembly. Contact-permitted operation requires measurement at the reachable contact points of the actual cell, with the actual tooling, at the actual programmed speeds.
What triggers a re-assessment?
Any change to speed, payload, tooling, layout, part family, operating type or safety-related software. A heavier part variant alone changes effective mass and therefore every derived speed limit.
Sources
- ISO 10218-1:2025, Robotics, Safety requirements, Part 1International Organization for Standardization
- ISO 12100, Safety of machinery, General principles for design, risk assessment and risk reductionThe general method underlying the sequence above
- Regulation (EU) 2023/1230 on machineryEU-OSHA, conformity duties and the 20 January 2027 application date
- Collaborative robot systems, information sheet 080German Social Accident Insurance (DGUV), measurement practice