The standard addresses a system, not an isolated vehicle
ISO 3691-4:2023 covers driverless industrial trucks and their systems, including AGVs and AMRs. A platform certificate is therefore not the whole safety case. The project must assess control, guidance, power, load handling and the actual operating environment. Public areas and certain special conditions fall outside this scope and require additional standards and risk analysis.
Before selecting a robot, document its purpose, load types, permitted speed, operating modes, reasonably foreseeable misuse and the responsibilities of the integrator, site owner and supplier.
The operating zone is part of the protective concept
Map pedestrian and vehicle routes, blind corners, crossings, doors, lifts, ramps, charging points and load-transfer stations. Define visibility, right of way, speed, audible and visual warnings, physical separation and the safe state for every critical location.
Inspect floor condition, gradients, clearances, lighting, reflective surfaces and temporary obstacles. A rack or aisle change after launch must be treated as a safety-system change rather than routine housekeeping.
Test normal operation and credible failures
Acceptance should exercise protective braking, holding on a slope, obstacle detection, communication loss, an incorrect load, low battery, manual and maintenance modes, automatic charging and recovery after a stop. Measure stopping distance at realistic mass and under the least favourable surface conditions.
Repeat tests at crossings, narrow passages and interfaces with gates, conveyors or lifts. Verify separately that fleet orchestration cannot issue conflicting missions and can safely remove a failed vehicle from service.
Operations must preserve the safety evidence
The operating pack should contain zone maps, risk assessment results, the list of safety functions, test records, manual-intervention rules, training and a change log. Updates to software, maps, sensors or fleet logic should trigger an agreed regression suite.
The practical readiness test is straightforward: an operator understands robot state, a nearby person can read its intent, and uncertainty drives the system into a verifiable safe condition.