AGV and AMR Risk Assessment: Standards, Zones, Stopping Distance and Who Is Responsible

The short answer

A mobile robot that comes with a declaration or a test report has been assessed as a product, for the operating environment its manufacturer specified. Your site is a different question: the routes, the racking, the people, the forklifts, the loads and the chargers are assessed when the robot is integrated, and again whenever they change. The general method is the one in ISO 12100: identify the hazards, estimate and evaluate each risk, reduce it, and repeat until the residual risk is acceptable. [1] [2]

Two families of documents structure that work for AGVs and AMRs. Internationally, ISO 3691-4 covers driverless industrial trucks and their systems. In the United States, the ANSI/A3 R15.08 series covers industrial mobile robots in three parts, for the manufacturer, the integrator and the user, while ANSI/ITSDF B56.5 covers line-guided AGVs. This article explains how they fit together. It is not the standard text, and it is not legal advice. [3] [4] [2]

Which document covers what

[1] [3] [5] [6] [4] [2] [7]

R15.08 groups robots by what they carry rather than by size: Type A is an AMR without attachments, Type B an AMR with a non-manipulator attachment such as a conveyor top, and Type C an AMR or AGV carrying a manipulator arm. A Type C robot also brings the arm’s hazards into the assessment. [4]

How AGVs and AMRs differ in routing and sensing is explained in how delivery robots navigate, AGV vs AMR.

What ISO 3691-4 leaves out

The exclusions decide whether you are inside the standard at all. ISO 3691-4:2023 does not cover trucks in public zones, for which it points to ISO 13482, nor operation on public roads. It also excludes trucks guided only by mechanical means such as rails, remotely controlled trucks, extreme climates, freezer use, strong magnetic fields, explosive atmospheres, and loads such as molten metal or radioactive material. Requirements for the power source itself are not covered. [3]

Two practical consequences follow. A sidewalk delivery robot is outside this standard. A fleet moved into a freezer store is outside its normal conditions, so cold-related hazards such as condensation on sensors, reduced battery capacity and slippery floors have to be assessed without its help. [3]

Three parties, three parts

[6] [4] [2] [8]

The Part 3 point most often missed is that the user’s duties restart whenever the user modifies the robot, the application or the operating environment after the supplier has left. Moving a rack, adding a conveyor, opening a new aisle or raising a speed limit in the fleet software are all changes of that kind. In practice the integrator often writes the first assessment, but the site owns it afterwards. [2]

Zones along the route

ISO 3691-4 asks for the operating zone to be prepared, and its Annex A classifies parts of the route into zones. According to TÜV Rheinland’s comparison with the older EN 1525, which had only a restricted zone and a hazard zone, ISO 3691-4 adds a confined zone and states clearer requirements for each, including truck speed. Zones are distinguished partly by whether the personnel detection means is active or muted. [3] [9]

The same summary cites clearance figures of more than 500 mm and a maximum of 0.3 m/s where personnel detection is muted. Treat those as pointers to the table to read in the standard, not as values to design to. What a site can do without the standard is the survey: walk every route and record, segment by segment, the side clearance between the vehicle and its load and any fixed structure. A narrow segment is where a person can be trapped between a moving vehicle and a wall or rack. [9]

Hazards to list for a warehouse route

[3] [9] [6]

TÜV’s summary reports that ISO 3691-4 lists 27 safety functions with a minimum required performance level each, with personnel detection and braking control at PLr d. A required performance level is a target for the safety-related control system under ISO 13849-1; it says nothing about whether a given vehicle meets it. That evidence comes from the manufacturer. [9]

Worked example: sizing a protective field

A laser scanner on the vehicle watches a protective field ahead of it. If a person enters that field, the vehicle must stop before it reaches them. The field therefore has to be at least as long as the distance travelled while the system reacts, plus the braking distance, plus an allowance for the scanner’s measurement tolerance. The scanner manual and the applicable standard define the exact formula and allowances; the structure below shows why speed dominates. [9]

Assume, for illustration only, a total reaction time of 0.14 s for the scanner and the drive, a constant braking deceleration of 2.5 m/s² measured with the heaviest load on the actual floor, and a 0.10 m tolerance. Reaction distance is speed × 0.14 s. Braking distance is speed² ÷ (2 × 2.5 m/s²).

Halving speed from 1.5 to roughly 0.75 m/s would cut the braking distance to a quarter, because braking grows with the square of speed. That is why fleets switch to lower speeds and shorter fields near docking faces, chargers and tight turns, where a 0.76 m field would keep seeing fixed structures and stop the vehicle. The same arithmetic shows what to measure. A worn brake, a wet floor or a heavier load lowers the deceleration, and every centimetre of braking distance it adds has to be inside the field.

The same reaction-time arithmetic for robots sharing space with people is worked through in why an edge vision model cannot set the safe speed.

What the assessment file should contain

• Scope: vehicle models, software and firmware versions, routes, loads, shifts and who works nearby.
• The hazard list, checked against ISO 3691-4 Annex B or the applicable R15.08 part.
• For each hazard: the estimated risk, the chosen measure, and the evidence that the measure works, such as stopping tests and the scanner field configurations with their version or checksum.
• Residual risks and what workers are told about them, with training records.
• A change log, and the triggers that reopen the assessment: a new route, a layout change, a new load type, a software or fleet update, an incident or near miss, and the periodic review.

[2] [8] [9]

The file is useful only if it matches the site as it is today. A walk of the routes with the field configurations in hand, comparing measured clearances with the ones recorded, is a quick way to find out whether it still does.

Limits of this article

Standards are paid documents and are revised: ISO 3691-4 is being replaced, and the third part of R15.08 appeared only weeks before this article. Requirements also depend on the jurisdiction. In the EU, Regulation (EU) 2023/1230 applies from 20 January 2027, and the conformity route depends on the product. Use this article to structure the questions, then answer them with the purchased standard texts, the manufacturer’s documentation and a qualified safety engineer. [5] [2] [7]

What a safety certificate does and does not prove is covered in robot safety standards and certification explained.