Autonomous Forklifts and AMR: Evaluating an Automation Project

The first quotation for a pair of autonomous forklifts nearly always arrives with the vehicle datasheet attached: capacity, lift height, speed, battery life. The decision needs three other documents, and no supplier can attach them. They are the movement history held in the WMS or ERP, a dimensioned floor plan with aisles, doors and slopes, and the calendar of peaks, which in many companies exists only in the shift supervisor's head and has never been written down.

With those three documents on the table you can answer the questions that matter for an investment of this kind: whether the flows are repetitive enough to support automation, how many vehicles the worst hour needs rather than the average one, which building constraints must be fixed before the order, and which clauses belong in the specification. Many of the problems that surface at start-up trace back to data nobody collected: an underestimated peak, a gate too narrow to pass at full speed, a floor that becomes a problem six metres up. The robot is usually the last thing to disappoint.

How many autonomous forklifts do you need? The worst hour decides

Before comparing an AMR with an AGV or an autonomous forklift truck, it pays to establish what has to move: which load units, between which points, how often and in which time windows. The mission map shows whether the process is predictable enough for automation, or still so variable that it needs a driver's flexibility. We set out the differences between the technologies in our comparison of AGVs and AMRs; the subject here is method.

A worked example, with round numbers, shows why the average misleads. A warehouse runs two shifts and moves an average of 30 pallets an hour between dock, storage and production lines, but in the first two hours of the morning shift the trucks arrive and missions climb to 55. If the study puts a full cycle of pick, travel, drop and return at six minutes, each vehicle completes at most ten missions an hour, before charging, waiting at gates and crossings are taken out. The average needs three vehicles. The peak needs six, twice as many, in the same building with the same technology. And a fleet sized on the average would still pass its acceptance test, if the test were run on a quiet afternoon.

That is why the movement history has to cover a period that actually contains the peak, ideally a full year of data logged by the system rather than estimated in a meeting. The economics are the ones the literature has long described for picking: the review by de Koster, Le-Duc and Roodbergen in the European Journal of Operational Research (2007) reports that order picking can account for up to 55% of a warehouse's operating cost, with travel as its heaviest component. Pallet transport is costed the same way. If travel time in your flows is small, or badly fragmented, no autonomous vehicle will win it back.

What data an AMR feasibility study needs

A serious supplier asks for numbers on the first call. It pays to arrive prepared, not least because almost everything is already in the company, scattered across the ERP, the floor plans and the memory of the people on the shop floor.

• Movement history: missions with time stamp, origin and destination, extracted from the WMS or ERP over a period that includes the seasonal peak.
• Hourly profile: missions per shift and, above all, the worst hour of the worst week.
• Load units: pallets, bins, carts, with dimensions, weights, stability and pick and drop heights.
• Dimensioned layout: aisle widths, slopes, level changes, doors, fire doors, goods lifts.
• Shifts and seasonality: operating hours and campaigns that change the calculation.
• Infrastructure: floor flatness, wireless coverage, power points for charging.
• Systems to integrate: the WMS, ERP or MES that will assign missions to the fleet and receive its status.

With these elements the conversation moves to a different level, because it stops being about the vehicle's qualities and starts counting how many missions are needed, in which window of the day and with what margin against the limit the fleet can carry.

What the feasibility study has to prove

The feasibility study turns this data into a decision that will stand up in front of the board. The German guideline VDI 2710 on the interdisciplinary design of automated guided vehicle systems, in its June 2025 edition, places it within a path that runs from system identification and detailed design through procurement and operation to change planning and decommissioning. The map is useful to buyers too, because it is a reminder that decisions taken at feasibility stage are paid for over the whole life of the installation.

A useful study proves at least three things. The first is capacity: how many missions the fleet completes in the peak hour, with a stated margin well short of the theoretical limit. The second is the number of vehicles, which depends on real cycle times, waiting at doors and crossings, and how much charging takes away from availability. The third is the economic return, read as total cost over several years; the method for building it, shifts and assumptions included, is in our analysis of ROI, TCO and payback of an AMR fleet.

There is also one thing no standard will write for you: the performance acceptance criterion. Safety standards say how the vehicle must protect people, and are silent on how many pallets it must move per shift. That figure is set by the buyer, and the study should hand it over already written in testable form, for instance missions per hour on a defined flow, measured in an agreed peak window.

When the numbers leave room for doubt, the prudent route is a pilot on a representative flow: a few vehicles, a real circuit, real data to set against the study. It only makes sense with an exit condition written before it starts, and valid both ways. If the pilot reaches the planned missions per hour on the chosen flow, it is extended; if it falls short, the layout is redesigned or the project stops, before the rest of the fleet has been bought.

Aisles, floors and network: the building constraints no catalogue shows

Many projects are decided by physical details. ISO 3691-4, the safety standard for driverless industrial trucks, devotes a normative annex to preparing the zones in which the vehicles operate: along the path it requires, on both sides, a clearance of at least 0.5 m in width and 2.1 m in height to adjacent fixed structures. Where that clearance is missing, or people cannot be detected, the area becomes an operating hazard zone, with reduced speed and additional audible or visual warnings. For the study this means something precise. Every stretch without clearance is a slow stretch, and a slow stretch on the busiest route eats missions per hour: the aisle, then, gets measured before the cycle time is estimated.

Floors matter most for trucks that lift high, and a little geometry shows why. A 5 mm level difference between the wheels of a vehicle with a one-metre track tilts the mast by five millimetres per metre: at six metres the forks sit about three centimetres off vertical, just as they have to slide a pallet into a rack bay. Wireless coverage has to be checked along every route, because a network gap is a stop, and doors, goods lifts and fire doors have to be built into the traffic logic. Charging, finally, is an availability variable and is sized together with the fleet, as explained in our piece on charging station sizing.

That leaves software integration, often treated as a detail. Vehicles must take orders from the management system and report their status, and with vehicles from different manufacturers interoperability becomes a long-term choice. The VDA 5050 specification, published by VDA and VDMA and at version 3.0.0 since March 2026, standardises the exchange of orders and states between a central control system and the vehicles. It has no legal force. It covers communication, while the safety conformity of the whole remains a separate matter, to be assigned by contract to someone who will sign for it. Asking for it in the specification costs one line, and keeps open the option of growing the fleet with another supplier.

EN ISO 3691-4:2023 and the Machinery Regulation: the standards to write into the specification

The safety of an autonomous handling system is settled at design stage. The technical reference is ISO 3691-4, which sets safety requirements and their verification for driverless industrial trucks and their systems, in its second edition since 2023. It took the place of EN 1525:1997, withdrawn in 2020 when the first edition of the ISO standard was published. Where CE marking is concerned, however, the form that counts is the harmonised European version, EN ISO 3691-4:2023, cited in the Official Journal by Commission Implementing Decision (EU) 2024/1329 of 13 May 2024: that is the version giving presumption of conformity with the Machinery Directive 2006/42/EC. Upstream sits risk assessment to ISO 12100:2010, and safety functions such as personnel detection, speed control and stopping are designed to the performance levels of ISO 13849-1:2023.

If a project evaluated today goes live after 20 January 2027, it will fall under Regulation (EU) 2023/1230, which replaces the Directive from that date. For the first time the Regulation deals with autonomous mobile machinery: it defines a supervisory function (Annex III, section 3.2.4), requires the control system to perform the safety functions by itself even when commands come through remote supervision, and requires movement to take account of the risks of the area where the machine works. The specification should therefore state under which regime the declaration of conformity will be issued, and who signs it for the complete system as well as for the individual vehicles. Groups with plants in the United States work to different references: ANSI/ITSDF B56.5-2024 for automatic guided vehicles and the R15.08 series for industrial mobile robots.

The weight of safety in a project, and its return in avoided injuries, is covered separately in our article on AMR safety.

The questions to put to a supplier before the quotation

Once suppliers are being compared, the questions weigh more than the datasheets. References are useful when they concern flows similar to yours. A list of installations, on its own, says little. Who takes charge of integration with the management system, and with what responsibilities, should be written down at once, to avoid the usual buck-passing between whoever supplies the vehicles and whoever runs the software. The number of vehicles proposed must be justified with your data and, where the route network is complex, with a simulation.

Then come service and spare parts, response times, the ability to add vehicles and, above all, who signs the conformity of the whole when the fleet mixes brands, a subject explored in our piece on automating pallet handling in a running warehouse. One signal deserves attention: a supplier who answers from the vehicle model rather than from your flow is often making a catalogue sale under a project's name.

Set-up mistakes that surface after acceptance

The most frequent stumbles concern method more than technology. The first is automating an inefficient process as it stands: the waste gets written into software and becomes harder to remove. The second is sizing on the average, as in the two-shift example, together with its twin, an acceptance test run in an ordinary week, which the system passes with ease and which says nothing about the November peak. The third is leaving operators out. Without their involvement start-up slows down, and the vehicles get bypassed by a few extra manned trucks, which spoils the statistics and safety at the same time. The last is choosing on purchase price, forgetting integration, service and cost over several years.

The opposite reading holds too. On a single shift, with few missions an hour and routes that change daily, autonomous vehicles spend much of their time parked or charging, and reorganising the manual flows first is often the better move. Saying so at feasibility stage costs little; finding out after the order costs a fleet.

If, on the other hand, the flows are repetitive and volumes stay high across several shifts, the next step is to put the movement history into a study that sizes the fleet on the peak and writes the acceptance criterion. The technologies that cover the different scenarios, from pallet transport to warehouse intralogistics, are described on the page dedicated to industrial automation and material-handling robotics; to start from your own data, PITECH sets up the assessment before proposing any vehicle.