Safety is not a document collected at the end of procurement. It is a design constraint that must be verified in the actual application.
A CE mark, a technical file, a brochure statement about obstacle avoidance and a demonstration in a clean showroom can all be useful pieces of information. None, alone, proves that a mobile-robot system will be safe in a particular warehouse or factory. Real safety performance depends on the machine, the load, the working environment, the interfaces and foreseeable human behavior.
The mistake is to treat safety as a procurement field marked “compliant.” The engineering task is to identify hazards, reduce risk and verify the resulting system throughout its lifecycle.
Begin with the actual application and its boundaries
Describe the intended use, robot class, maximum and typical loads, routes, speeds, gradients, working hours, operating modes, access restrictions and interacting equipment. Include manual operation, maintenance, fault recovery and foreseeable misuse where relevant.
A low-profile under-cart AMR operating in a segregated warehouse and an autonomous forklift raising a heavy load in mixed traffic present different hazards. A generic declaration that “the robot is safe around people” is not an adequate assessment of either application.
Define the system boundary clearly: does the risk assessment cover the vehicle only, or also its docks, doors, lifts, conveyors, transfer stations, fleet controls and site traffic arrangements? Gaps between supplier and integrator scopes are often where responsibilities become unclear.
Use a structured risk-reduction method
The general framework of ISO 12100:2010 is useful for identifying hazards, estimating and evaluating risk, and applying risk-reduction measures. For driverless industrial trucks and their systems, ISO 3691-4:2023 is a relevant type-specific safety standard; applicability depends on the equipment and the jurisdiction. Other standards and legal requirements may apply to a particular solution.
Consider collision and crushing hazards, trapping at transfer points, unstable loads, uncontrolled movement on slopes, restricted visibility, falling loads, loss of localization, recovery actions and human entry into operating zones. Risk cannot be evaluated responsibly from a floor plan alone. Site visits and actual task observation matter.
Apply the established hierarchy: first seek inherently safer design choices, then appropriate protective measures, and finally information for use and training for residual risks. Signage and operator training should not be used as substitutes for practicable engineering safeguards.
Distinguish obstacle detection from a validated safety function
A robot can have sophisticated perception without every sensor or software behavior being a safety-rated protective function. Ask which functions are designed and validated for risk reduction: protective stopping, speed control, detection fields, interlocks and relevant operating modes. Request evidence of their performance and operating limits.
Test under relevant scenarios and configurations. Safe stopping depends on speed, payload, gradient, tire/floor friction, reaction times and the protective system's design. A test result with an empty robot on a clean floor may not represent a heavily loaded vehicle approaching a crossing.
Robot behavior during localization loss, protective field obstruction, communication failure or an emergency stop must be defined and validated. An emergency-stop button is necessary in many contexts but is not a complete safety concept.
Mixed traffic changes the risk picture
Forklifts, pedestrians, hand trolleys and AMRs require deliberate traffic engineering. Review shared aisles, pedestrian crossings, blind corners, parking areas, temporary storage and the places where staff are likely to intervene with a blocked robot.
Segregation, controlled crossings, barriers, speed zoning, visibility measures and operating rules may be appropriate depending on the application. Their effectiveness must be verified. Consider whether a robot's chosen path leaves enough clearance to avoid trapping a person against fixed equipment.
Do not assume that a “collaborative” marketing description means a machine can operate without a site-specific assessment. Safe human–robot interaction is an outcome of the complete application design, not a product label.
Make conformity and responsibilities explicit
For EU projects, the legal framework and conformity obligations must be determined for the equipment, date of placing on the market, integration scope and possible substantial modifications. As of November 2026, Machinery Directive 2006/42/EC remains relevant to machinery placed on the EU market before the new Machinery Regulation (EU) 2023/1230 applies generally on 20 January 2027. The required conformity route cannot be inferred from a checklist alone.
Ask for applicable declarations, technical documentation, instructions, test evidence and the risk assessment scope. Confirm who is responsible for the final integrated system, modifications, site safety arrangements and acceptance. Supplier documentation is essential evidence, but it does not remove duties arising from local integration or use.
Validate during FAT, SAT and commissioning
Plan tests before the robot arrives. Define conditions, measurable criteria, safe test procedures and who signs results. FAT may establish selected functions in a controlled environment; SAT must assess the system against actual site interfaces and operating conditions. A successful FAT does not eliminate the need to verify the real installation.
Test representative protective behavior, handover areas, alarms, manual recovery and changes of operating mode. Include operator and maintenance training, access-control responsibilities and documented procedures for clearing faults. Changes to routes, loads, sensors or transfer stations should trigger a review of their safety impact.
Safety evidence checklist
- Defined intended use, operating envelope and system boundaries
- Application-specific hazard identification and risk assessment
- Safety functions and limitations supported by evidence
- Mixed-traffic and transfer-zone safeguards evaluated
- Applicable standards and conformity obligations confirmed
- FAT/SAT test conditions and acceptance criteria approved
- Residual-risk instructions, training and recovery procedures in place
- Change-control process maintains the safety case after go-live
The engineering conclusion
Safety is not proven by one certificate, one sensor or one successful demonstration. It is established through documented risk reduction, appropriate system design, evidence-based validation and disciplined operation. The most reliable project is the one that takes hazards seriously before they appear in daily production.
Technical note: This article offers an engineering approach, not a product-specific legal conformity assessment. Confirm the current legal and standards requirements for the machine and market concerned.
Further reading: ISO 12100:2010 · ISO 3691-4:2023 · European Commission — Machinery.
ČAMINSKI — Engineering Advisory & Project Delivery. For independent safety-readiness and acceptance-scope reviews: office@dimitarcaminski.com.