AGV Layout Design 2026 | Fleet Size Is Not Set by Throughput

“How many AGVs do we need?” is where almost every conversation about automating in-plant transport begins. It is also the question that cannot be answered yet. Fleet size is not set by throughput alone. The time a vehicle spends charging, the time it spends waiting for another vehicle at an intersection, the time it spends stopped because a pallet is sticking out into the aisle — all of that is decided by the place the vehicles run in, and none of it appears in a vehicle catalogue. This article treats AGV layout design as the work that comes before choosing a vehicle, and sets out five things — routes, floor, stopping accuracy, charging and host system integration — in the order the buying side should decide them.

A note on the figures used in this article. Every statistic is tagged so you can tell whether it transfers to your own site. Global marks an international standard or a worldwide market estimate. Thailand-specific marks something that applies to operations in Thailand and should not be carried across to Vietnam, Indonesia or the Philippines without checking the local equivalent. Japan-specific marks a figure that comes from a Japanese source and describes Japanese practice or the capability of a Japanese contractor.

Why “how many do we need” cannot be answered yet

The first questions that arrive from a plant starting to look at transport automation are, almost without exception, fleet size and price. How many vehicles for a process like ours? What does one cost? It is natural to assume that once those two are answered a capex request can be written. In practice those two are precisely the items that cannot be answered while the conditions of the place the vehicles will run in remain undefined.

Quotations miss for reasons that have nothing to do with vehicle selection

Look back at projects where the outcome was “it does not move as much as we expected” or “it cost more than the quotation”, and the cause is rarely the model of vehicle that was chosen. What comes up far more often is this sort of thing.

• The aisle turned out to be too narrow, so the two way running originally planned was changed to one way. Travel distance grew and the time for one lap went past the assumption
• Floor unevenness was worse than assumed, so a speed limit was set for that stretch. The designed cycle time no longer held
• All the chargers were installed in one place, and a queue formed in front of them. Adding vehicles did not add throughput
• Host system integration was left out of the first phase and replaced with manual entry, so when automatic integration was added in phase two, modification cost appeared
• Safety devices had to be added at intersections, and electrical work turned up as a line item that had not been in the original quotation

What these five have in common is that none of them is about vehicle performance. Every one of them is a condition of the place the vehicles run in. And every one of them becomes an extra cost when it is discovered after the vehicle has been selected. Turn it around: if the buying side defines these on a drawing before selection, they are built into the premise of the quotation from the start.

AGV layout design is not something you do after choosing the vehicle

The usual mental picture of how this goes is probably something like the following. First count the transport volume. Then compare makers and select a vehicle. Finally draw the routes to suit that vehicle. As a sequence it looks plausible enough.

There is a fundamental problem with it. As the later sections work through in detail, the fleet sizing formula itself depends on the layout. How many intersections there are, where the chargers sit and how many of them, whether an aisle is one way or two way — each of these changes how many runs one vehicle can complete in a day. Fleet size is nothing more than the result of that division. A fleet size produced while the layout is still undefined is a number that has been fixed while its basis is still blank.

So the claim of this article reduces to a single sentence.

AGV quotations and utilisation figures miss not because the wrong vehicle was selected, but because the fleet was counted before the buying side defined the conditions of the place it would run in.

What this article does not cover

By way of scope: the difference between AGVs and AMRs, the classification of machine types by guidance method, and the prior question of whether automated transport is the right answer at all are not covered here. Those are set out in how to judge an AGV or AMR project and the order of magnitude of the cost, and this article picks up at the next stage, where the decision to proceed has been taken and the buying side has to define where the vehicles will run and how.

The reader we have in mind sits in production engineering, plant engineering, maintenance or plant IT, and has been handed a request from inside the business to automate transport and turn it into a specification. Rather than handing the whole thing to a vendor, the aim here is to work through what the buying side can settle in advance so that the quotation is more accurate and there is less to regret once the system is live, in the order those decisions should be made.

Fleet size is not set by throughput — the three time sinks a naive calculation ignores

Here is the central claim of the article, stated up front. AGV fleet size is not a linear function of throughput.

The naive formula and its hidden assumption

The first formula that appears when anyone estimates fleet size is roughly this.

Fleet size = (transport runs per day × cycle time per run) ÷ available operating time

There is nothing wrong with the formula itself. What is wrong is treating the “available operating time” on the right hand side as identical to the plant’s operating hours. The time an AGV actually has available for transport in a day is what remains after three separate blocks of time have been taken out of the operating hours.

Those three are worth taking one at a time.

Missing time one — charging occupancy

While a vehicle is charging it is not transporting anything. Obvious enough, and yet it is the easiest item to drop out of the denominator.

Charging affects utilisation through three separate channels.

First, the charging itself. The approach differs by battery type. Some operations run continuously with repeated short top ups; others take the vehicle out for a full charge over a longer block of time. Either way, that time is not available for transport.

Second, the round trip to the charger. This is where it becomes a layout question. Grouping the chargers in one place keeps equipment cost and electrical work down, but the journey from the far end of the route back to the charger is entirely non transporting time. Distributing them along the route shortens the round trip but multiplies the number of places needing an electrical supply. Charger placement is a design variable that trades cost directly against cycle time.

Third, waiting to charge. If the number of chargers is short relative to the number of vehicles, a queue forms. This is the most easily overlooked of the three. If a vehicle is added and a charger is not, the transport capacity of that added vehicle is not obtained in full.

Missing time two — waiting at intersections

Of the three, this has the largest effect and is the least likely to be reflected in a quotation.

When several AGVs share one aisle network, one of them has to wait at every intersection, because the master control system stops one of them to prevent a collision. That waiting time grows on a per vehicle basis as the fleet grows, because more vehicles passing through the same intersection means more mutual waiting events there.

An important consequence follows. There is a region in which adding vehicles does not raise total throughput as much as expected. Past a certain point, the increase in waiting time that an added vehicle imposes on the existing vehicles can outweigh the transport capacity it brings. “We added AGVs and the floor did not get any easier” is, in most cases, this effect.

It would be wrong to put a number on it. Where the ceiling sits depends on the shape of the aisle network, the number and position of intersections, the distribution of destinations and the algorithm the master control system uses. There is no general answer of the form “past a given number of vehicles efficiency drops”, and where a document appears to state one, it is an observation under one particular set of conditions.

What has practical value is not memorising a threshold but having the behaviour simulated on your own aisle network. The section on what to demand from a system integrator makes that concrete.

Missing time three — exception stops

The third block is time spent stopped because of something nobody planned for.

• A person is standing in the aisle, or crouched down working in it
• A pallet or an empty crate is protruding into the aisle
• The vehicle has met a hand trolley or a forklift head on
• The vehicle has lost its position and needs to be returned to a reference point
• Recovery work after a power outage or a voltage dip

These are not things to be eliminated. They are things to assume will occur at some frequency and subtract from the denominator in advance. A factory is a place where people work, and there will not be a day when nothing is left standing in an aisle.

Sites in Thailand carry one additional local factor. Voltage dips during the rainy season are Thailand-specific. If the voltage drops for even an instant and the master control server or the on board controller restarts, position information is lost and a return to reference is needed. Each individual recovery may be short, but if dips are frequent through the rainy season, the monthly total is not negligible. Who performs that recovery, and what happens if it occurs during unattended night running, are questions that belong at the same stage as the layout design.

Which is why fleet sizing only works after the layout

Put the three blocks back in and the earlier formula reads as follows.

Fleet size = (transport runs per day × cycle time per run) ÷ (operating hours minus charging occupancy minus intersection waiting minus exception stop time)

And all three of the terms subtracted in that denominator are set by the layout.

As long as the denominator is a dependent variable of the layout, it is not possible in principle to produce a fleet size without deciding the layout. That is the argument behind the claim that AGV layout design is not something you do after choosing the vehicle.

What to demand from a system integrator at quotation stage

Given all of the above, what the buying side should do when requesting quotations becomes clear. Rather than specifying a fleet size and asking for a price against it, hand over the conditions, ask the integrator to propose a fleet size, and require the basis for it as a deliverable.

Specifically, ask for these three things.

1. The result of an operating simulation that includes intersections. Not a simple round trip distance calculation, but a demonstration that the proposed fleet meets the required throughput once mutual waiting at intersections is accounted for
2. The charging plan. The number and installed position of chargers, the number of charges per vehicle per day and the occupancy time of each, and why no charging queue forms
3. How exception stops have been allowed for. What utilisation percentage the calculation assumes, and on what basis

A quotation that arrives without these three has a blank where the basis for its fleet size should be. Conversely, an integrator who can produce all three has been thinking about intersections and charging from the design stage. In practice this works as a genuine selection filter when comparing quotations.

If you are still at the stage of deciding which of several sites or processes to automate first, that prioritisation is faster to settle upstream, before any single transport project is scoped in this level of detail.

The five things the buying side decides first, and the order they go in

From here the article organises what the buying side should commit to drawings and documents into five items. The order matters, because each later item is a dependent variable of the ones before it. Reverse the order and the same decisions get made repeatedly.

The row to pay most attention to is the second one, the floor. The floor is the only item that, once running has begun, cannot be corrected without stopping production in that area. A route change on a drawing is an eraser; grinding or recoating a floor is construction work. So the floor gets measured immediately after the candidate routes are drawn, and before a vehicle is selected.

The five follow in order below.

Routes and aisle width — read occupied width plus safety clearance plus passing, not vehicle width

The first thing to decide is where, in which direction, and how many vehicles will run.

Do not judge aisle width by vehicle width

A maker’s catalogue gives the overall width of the vehicle. Use that figure directly to judge aisle width and it will almost certainly come up short. The width actually needed has to be treated as the sum of at least three components.

First, the occupied width of the vehicle. Not just the body width, but any overhang of what it is carrying. Where a pallet is larger than the body, where long items are loaded, where a trolley is being towed, occupied width goes well beyond body width. On a curve, the difference between the paths of the inner and outer wheels means the vehicle occupies more width than it does running straight. The width needed on a curve is greater than on a straight, which is obvious enough and still gets missed on the drawing.

Second, safety clearance. The distance required between the side of the vehicle and walls, racking, columns and equipment. It has to cover the detection range of the sensors and the deceleration and stopping behaviour that follows, plus an allowance for deviation. Squeeze this too hard and the vehicle starts stopping whenever something is temporarily set down beside the aisle, which increases exception stop time.

Third, allowance for passing. On any stretch that is to be two way, the width has to let two vehicles pass. Where that cannot be achieved, the stretch either becomes one way or needs a pull off bay.

Reading it as aisle width = occupied width + safety clearance + (if passing, the occupied width and clearance of the other vehicle) makes it clear that the catalogue width and the width actually required are entirely different numbers.

One way or two way is decided by intersection count, not by aisle width

Deciding to make a stretch one way because the passing width is not available looks like a natural judgement. The consequences of that judgement do not stop at aisle width.

Going one way means the route becomes a loop. A loop lengthens travel. Returning to a point already passed means going all the way round. Longer travel means a longer cycle time per run, which enlarges the numerator of the fleet size calculation.

Two way running shortens the distance, but oncoming vehicles have to be arbitrated, pull off bays are needed so they can wait for each other, and intersection management becomes more complex.

Real factories usually end up on the third row: main arteries two way, narrow branches one way. The design point in that case is to place a pull off bay at every boundary where the scheme changes.

Intersection count can be reduced by design

As set out earlier, the number of intersections governs cycle time through waiting. And the number of intersections can be reduced by how the routes are drawn.

There are three ways of thinking about reducing them.

Do not overlay routes. If several transport jobs — line replenishment, empty crate collection, finished goods removal — can be given separate paths, they do not cross. Even where the plant layout makes sharing an aisle look natural, separation by time of day or by zone is sometimes available.

Replace crossings with merges and branches. A four way crossing requires arbitration between four directions, whereas connecting loops to each other decomposes the same traffic into a sequence of merge points and branch points. The complexity of arbitration at each location drops.

Put intersections where work density is low. The area around an intersection becomes a place where vehicles stop and wait. If that overlaps a walking route or a temporary set down area, exception stops concentrate there. Deliberately offsetting intersections from the areas where people work is something that can only be done at the layout stage.

Design for coexistence with people, hand trolleys and forklifts

Unlike a purpose built distribution centre, manufacturing sites in Thailand and across ASEAN rarely allow an aisle to be reserved for AGVs alone. The vehicles will run in aisles where people walk, hand trolleys pass and forklifts cross.

On that basis, the design decisions to make are these.

• Which stretches are separated and which are shared. Separating everything is not realistic, so the stretches need to be ranked
• Priority on shared stretches. Does the AGV stop and wait for people, or do people step aside for the AGV? This is an operating rule and at the same time it feeds directly into the deceleration settings on the vehicle
• Where temporary set down is prohibited. Keeping the sides of an aisle clear is in practice the hardest thing to sustain in daily operation. Painting a line on the floor does not on its own keep it clear. Unless a separate place to put things is provided, things end up in the aisle
• Notices in the languages actually spoken. On sites in Thailand it is not unusual for the workforce to include people whose first language is Burmese, Khmer or Lao alongside Thai speakers, and comparable mixes occur elsewhere in ASEAN. The languages in which AGV running rules are posted need to match the people actually working there (Thailand-specific in its detail, though the underlying issue is regional)

The last two are easy to dismiss as not being equipment questions, and yet they are exactly what determines exception stop time. As measures that protect the denominator, they deserve the same weight as the layout design itself.

The floor is a cycle time problem before it is a safety problem

Of the five, the floor is the only one that stops production if it is corrected later. That is why it gets its own section.

Even slight unevenness forces a slowdown

Floor Agent, a Japanese flooring contractor, notes of AGV running conditions that even slight unevenness interferes with running and forces a reduction in speed (Japan-specific, from a Japanese contractor’s technical column). That one sentence points at the most important way of looking at the floor.

Frame the floor as a question of whether it is dangerous, and a certain amount of irregularity looks acceptable. What actually happens is different: the speed setting for that stretch gets reduced. And a lower speed means a longer cycle time per run, which enlarges the numerator of the formula in the fleet sizing section and increases the number of vehicles required. The floor is a cycle time and fleet size problem before it is a safety problem.

The same contractor identifies three floor symptoms as problematic.

The remedial work identified is levelling work such as grinding and floor coating.

The finish tolerance a contractor quotes is in the single digit millimetre range

The same contractor states that floor level error can be finished to an accuracy of the order of 3 mm to 5 mm (Japan-specific, from a Japanese contractor’s technical column).

Two cautions apply.

First, this is not a standard. It is one example of the finish tolerance a contractor offers. The figure should not be treated as a public required value of the form “3 mm to 5 mm is the benchmark”. It is information about the order of magnitude that levelling work can reach, and nothing more.

Second, the interval over which that error is measured is reported inconsistently across sources. Flatness is a metric that only carries meaning together with the distance or area over which the difference is taken, and quoting a bare number while the definition of that interval is unsettled makes for a shaky basis when building internal agreement. This article quotes only the point that the error is in the single digit millimetre range, and does not assert the interval unit.

The formal flatness requirement comes from the AGV maker’s specification

So what is the right instrument for judging whether your floor meets the requirement? The answer is clear. It is the specification issued by the maker of the vehicle you are considering (global, in the sense that this holds regardless of country).

Flatness requirements vary with wheel diameter and count, the presence or absence of suspension, the guidance method, running speed and payload. On the same floor in the same factory, vehicle A may be fine while vehicle B needs to slow down. A pass or fail judgement on the floor therefore only works in one sequence: measure your own floor first, then obtain the maker’s required values once the vehicle candidates have been narrowed, and compare the two.

In practice, for the buying side, that looks like this.

1. Draw candidate routes on the drawing
2. Measure the current floor condition along those candidate routes (unevenness, steps, joints, gradient, surface material, condition of any existing coating)
3. Obtain flatness requirement specifications from the makers of the candidate vehicles
4. Compare measured values against required values and identify the stretches that fall short
5. Include the scope, method, cost and duration of work on those stretches in the quotation

Do these five steps before requesting quotations and floor works stop being a cost that appears afterwards. Place an order without doing them and the floor problem surfaces at installation, leaving a choice between stopping production for the work or compromising with a reduced speed setting.

Several documents cite specific standard numbers for floor flatness. This article takes the position of not quoting material whose content was not verified during this research. If a standard number is to be used as the basis for a decision, it should be checked with the vehicle maker that the standard actually corresponds to that vehicle’s requirement.

Joints, steps, gradients and surface materials get the same treatment

Beyond unevenness, several other floor conditions affect running.

• Joints (expansion joints and induced crack joints). Where the width and depth of the groove are large relative to the wheel diameter, every crossing produces vibration and a slowdown. And where guidance relies on something applied to the floor, such as magnetic tape or two dimensional codes, the position of the joints becomes a constraint on where it can be applied
• Steps. Boundaries between zones, door frames, drainage gratings, the transition at the foot of a ramp. On a ramp it is not only the gradient that matters but the break points at the start and end of the climb, where the underside of the body can make contact or a drive wheel can lift off
• Gradient. Floors with a deliberate fall for drainage are common. It affects both climbing performance and braking distance on the descent
• Surface material. Type of coating, presence of dust proofing, bare concrete, steel plate. A change in friction coefficient changes braking distance and acceleration performance, and where the material changes between stretches, behaviour changes at the boundary
• Water, oil and dust. Around washing processes, in zones where cutting oil is thrown off, in areas handling powders. Where the surface gets contaminated, running is not stable even on a flat floor

All of these can be measured once candidate routes exist. Without a route, there is no way of knowing where to measure. That is why the order is routes first, floor second.

Plan floor works together with the installation window

Anyone planning floor works at a site in Thailand runs into the timing question immediately. Grinding and coating need curing time, during which the area is unusable. In a running factory that means the work can only be done during a substantial shutdown.

The installation windows realistically available in Thailand are Songkran in April, the year end and new year period, and the days around Chinese New Year (Thailand-specific; other ASEAN countries have their own equivalents, such as Tet in Vietnam). Miss one and the next opportunity is months away.

So the planning works backwards.

• Fix the floor works schedule against the installation window
• Work back from there to set the deadline for measuring the floor and settling the method
• Work back again to set the deadline for narrowing the vehicle candidates and obtaining maker specifications
• Work back once more to set the deadline for finalising the route drawings

When an AGV go live date is being set, what actually governs it is often not vehicle lead time but the floor works window. Establishing that at the start of planning avoids committing to a schedule that cannot be met.

Stopping and transfer point accuracy — the process decides how tight it has to be

The third item is the accuracy of the stopping position and of the load handover.

The process sets the accuracy requirement, and the requirement sets the vehicle options

“How accurately does it need to stop?” is not, properly speaking, a question for the maker. It is a question the process side has to answer.

• Transferring onto a conveyor means the receiving width of the conveyor and the shape of its guides set the requirement
• Docking at the input and output station of an automated warehouse means the tolerance of the station sets it
• A robot picking a workpiece off the AGV means the permissible deviation against the robot’s taught position sets it
• A person receiving the load by hand relaxes the requirement considerably

And the tighter the accuracy requirement, the fewer the guidance methods and transfer mechanisms available. That is why item three belongs early. Tightening a stopping accuracy requirement after the fact can send the project back to reselecting the vehicle itself.

Put the accuracy in the vehicle, or put it in the receiving side

In practice there are two ways to solve an accuracy problem.

Approach A: raise the stopping accuracy of the vehicle. Adopt a higher accuracy guidance method and give the vehicle a position correction mechanism. The design is straightforward, but unit price rises and the management of floor surfaces and markers becomes more demanding.

Approach B: put a positioning mechanism on the receiving side. Assume some scatter in the stopping position and provide guides, tapers and centring mechanisms at the receiving end. Because the requirement on the vehicle can be relaxed, the range of usable vehicles widens.

The third row is the one to look at. A configuration that puts accuracy in the vehicle means buying another expensive vehicle every time the fleet grows. A configuration that absorbs it at the receiving side, with a fixed number of stations, adds only the cost of the vehicle itself. Where fleet growth is expected, putting the accuracy in the receiving side tends to be better on total cost. Where the number of stations is large, the comparison can invert, so both need to be costed and compared.

Where the counterpart at the transfer point is a robot, this judgement connects directly to the robot’s teaching. The design thinking behind processes where a robot handles loads is set out in how to run a palletising robot project, which is useful background when thinking about the connection point with an AGV.

Transfer points are also the connection to storage

Transfer point design is not purely a transport question. In most cases the origin or destination of an AGV run is a piece of storage equipment: a pick from racking, an input or output station of an automated warehouse, a buffer area for work in progress.

The point to hold onto is that the throughput of the storage equipment can become the ceiling on the throughput of the AGV fleet. Add vehicles, and if there is only one input and output station, a queue forms there. The same structural problem as waiting at an intersection appears at the transfer point.

How storage side configurations and their costs work is covered in how automated warehouse pricing and configuration work. Where transport and storage are studied separately and connected afterwards, this capacity mismatch tends to surface late.

The list to write down before ordering

The items to document about stopping and transfer points before placing an order are as follows.

• Required stopping accuracy at each transfer point (along the direction of travel, across it, and in angle)
• Transfer method (conveyor, lift, rollers, robot, manual)
• Time taken by the transfer (this time lands directly in the cycle time per run)
• Space the vehicle occupies during the transfer (does it block the aisle, or can it pull off)
• Variation in load presentation (pallet types, dimensional tolerance, weight range, centre of gravity)
• Behaviour when the receiving side is full or empty

The last item is the one that goes missing. When the receiving side is full and the load cannot be set down, does the vehicle wait where it is, go somewhere else, or call a person? If that is undecided, a vehicle stops in the middle of an aisle on the first day of operation.

Charging and battery operation — charger position is a layout design variable

The fourth item is charging. As set out earlier, of the terms in the denominator of the fleet size calculation, this is the one the buying side can design most directly.

Where you charge decides how much you can move

The choice of charging method sets the shape of the operation.

Which is right depends on the transport pattern, but the two things to look at from a layout point of view are the distance travelled in order to charge and the probability of waiting in front of a charger. The configuration that minimises both is the configuration that moves more with the same number of vehicles.

The important corollary is that the position of a charger is also the position of electrical work. A factory’s electrical distribution is already laid out, and every additional place needing a supply raises the cost of the work. The trade off between “distributing chargers gives a better cycle time but costs more in electrical work” can be turned into numbers at the layout design stage. Argue it after the vehicle has been selected and it becomes a compromise with one side of the equation already fixed.

Fold the Thai electricity tariff structure into the charging plan

Industrial electricity in Thailand includes a tariff structure with different unit rates for on peak and off peak periods (Thailand-specific; other ASEAN countries have their own tariff structures). Where the timing of charging can be controlled to some extent, shifting it towards off peak can reduce part of the running cost.

That assumes the periods of high transport demand and the periods when charging is wanted do not conflict. In something close to continuous operation, there is little room to defer charging at all. Delaying charging to reduce the electricity bill and then finding that daytime transport does not keep up defeats the purpose.

A realistic sequence for working through it is this.

1. Build a charging plan that meets the transport requirement, without regard to time of day
2. Within that plan, identify the charges that have room to move in time
3. Shift those, and only those, towards off peak

Treat electricity cost optimisation as a dependent variable of the transport plan. Reversing the two leads to trouble.

Batteries are consumables

The item most often missed in a quotation comparison is battery replacement cost. Battery capacity falls with charge and discharge cycles. Lower capacity means less distance per charge, more frequent charging, and less time available for work. A fleet size calculation built on as-new specifications may not hold several years later.

The items to confirm before ordering are these.

• The expected replacement interval (in charge and discharge cycles, or in years)
• The cost of replacement, and what that cost includes
• Who performs the replacement (in house maintenance, or a vendor visit)
• The supply route and lead time for replacement parts (whether stock is held in Thailand or the parts are imported)
• How operation with degraded capacity is handled (whether the fleet carries margin, or replacement is brought forward)

The fourth item matters particularly at sites in Thailand and elsewhere in ASEAN. If parts are not held in local stock, the lead time includes import and customs clearance. Where that is the case, the operating plan needs to state how the period without transport capacity will be covered.

Think of charger count as fleet size plus a margin

The charging queue problem raised earlier is worth stating explicitly as a design item.

When a vehicle is added, it is worth confirming that a charging slot for it has been added as well. Where the number of chargers stays fixed and only the fleet grows, part of the added vehicle’s available time is consumed waiting to charge.

A useful way to write this into the specification is as follows.

• The proposal is to state a charger count at which no charging queue occurs for the assumed fleet size
• The proposal is to state alongside it the charger count and electrical capacity needed if the fleet is later expanded to a stated number
• The proposal is to show the charging plan behind those figures (charges per vehicle per day and occupancy time per charge)

The second line is the one that pays. Confirming the future electrical capacity while still in the first phase avoids a situation in phase two where the work has to start back at the incoming supply. Electrical work is a textbook case of a cost that is cheap when settled during layout design and expensive when added later.

Writing host system integration into the phase one specification — WMS, MES, lifts and shutters

The fifth item is integration with host systems. Whether this was written into the phase one specification is what sets the cost of expanding in phase two.

Terminology is worth pinning down before going further, because two different things are easily conflated. The master control system is the software that dispatches and coordinates the vehicles. The host systems are WMS, MES and production control, which sit above it and decide what needs to move and when. They are different layers with different interfaces, and the rest of this article keeps them apart.

“Manual for now” comes back in phase two

In a first phase it is common to leave host integration out and have transport instructions typed into a tablet. Development cost is contained and start up is quicker. There are situations where that is a reasonable judgement.

The problem is what comes after. As transport spreads to a second and third process, the labour of manual entry accumulates. And when automatic integration is then added, the following tends to happen.

• The master control system has no external interface, and modification is required
• The data structures in the host systems (WMS, MES, production control) do not support transport instructions, so those need modification too
• Operating rules built around the phase one way of working do not fit automation
• The boundary of responsibility between the vendor who supplied phase one and the vendor who looks after the host systems was never settled

The result is that development which would have been straightforward at the outset turns into modification and coordination cost.

So the recommended way to write it is this. The first phase does not have to implement automatic integration. What it does have to do is state, in the phase one specification, the interface specification for the automatic integration that will come later. Concretely, require the proposal to state which data is exchanged in which direction, which communication method is used, and that expanding later will not require modification of the master control system.

What to decide for each counterpart

Lifts and doors are the line items most often absent at quotation stage. Calling an existing lift from an AGV usually requires modification on the lift side to accept an external signal. That is work performed by the lift maker, and it is not inside the AGV vendor’s quotation. The answer to “can the AGV move between floors?” is set by whether the lift can be modified, not by the capability of the vehicle.

The same applies to fire shutters and dust control doors, where the work lands in the control panel on the door side. Modification of existing equipment of this kind involves PLC changes and signal interfacing. The questions around who owns how much of the equipment side control are the same ones covered in outsourcing PLC program development, and they transfer to this situation directly.

The buying side decides the trigger condition for a transport order

The hardest thing to design in host integration is not the communication method. It is when transport starts.

• Move when a set quantity of finished goods has accumulated
• Move when stock at the next process falls below a set level
• Move at fixed times based on the production plan
• Move when an operator presses a button

This choice changes the number of transport runs directly, and therefore changes the fleet size. It is also a matter of how the process is run, which is not something a vendor can decide. The buying side has to settle it.

In practice, observe when transport is triggered today, then decide first whether to automate that as it stands or to change the way the process is run while the opportunity is there. Automating the current way of working is the safest option, but where the current way of working is inefficient, the inefficiency gets automated along with it. The trade off is worth choosing consciously rather than by default.

VDA 5050 version 3.0.0 changes how the specification is written (published March 2026)

On the communication between vehicles and the master control system, there is a new consideration as of 2026.

What has been standardised

Version 3.0.0 of VDA 5050 was published in March 2026 (global).

VDA 5050 is a specification developed jointly by the German Association of the Automotive Industry (VDA) and VDMA e.V., with support from the Institute for Material Handling and Logistics at the Karlsruhe Institute of Technology (KIT IFL). By definition it is a communication interface between the master control system and mobile robots, with intralogistics and Industry 4.0 as its target domain.

It is worth being precise about the scope. VDA 5050 is an interface between the master control system and the vehicles. It is not a standard for connecting to WMS or MES. Host system integration, covered in the previous section, remains a separate piece of work with its own interfaces.

The technical basis is described as JSON messages exchanged over MQTT, with a single broker serving multiple clients, meaning the vehicles. This description of the technical arrangement rests on secondary sources, so anyone using it as a premise for implementation is advised to confirm it against the specification document itself.

Improvement proposals are accepted through a GitHub repository, and VDA and VDMA review them for incorporation into future versions.

What changes for the specification document

Where this interface is supported, vehicles from different makers can be run together under a single master control system. That is what it means in practice.

The traditional picture was this. A plant installs AGVs from maker A, so the master control system is also maker A’s. When the second phase extends into another zone, even if maker B’s vehicles suit the conditions better, a separate master control system would be needed, so in effect maker A keeps being chosen. The choice of maker in the first phase locked in the options for years afterwards.

A configuration built on a standard interface loosens that lock-in. That changes what belongs in the specification document.

• Whether the proposed master control system supports the standard interface
• If it does, which version
• If it does not, whether there is a plan to support it
• The support status on the vehicle side (mixed operation is impossible if the master control supports it and the vehicles do not)
• Whether there is evidence of mixed operation having actually been verified

The fifth line is the one that matters in practice. Claiming conformity to the specification is a different thing from having actually run vehicles from different makers under one master control system. Check operating evidence rather than a compatibility table in a catalogue.

Be careful with versions

The official site states explicitly that older versions of VDA 5050 are no longer recommended and will not be developed further (global). Older versions remain obtainable from the archive, but they are not recommended for new adoption.

So a specification document needs to state the version rather than simply saying VDA 5050 is supported, because products supporting only an older version can exist in the market.

Conformity does not equal mixed fleet operation

One qualification, by way of setting expectations. A standard interface aligns the communication between the master control system and the vehicles. It does not by itself guarantee that vehicles from different makers will run together without problems.

In reality, vehicles differ in dimensions, turning radius, acceleration and braking performance, and the detection range of their safety sensors. Running them in the same aisles means aisle width and intersections have to be designed around the most demanding of them. In other words, if mixed operation is the intention, that intention has to be built into the layout design. Here too, conformity to a standard is not a reason to leave the layout until later.

What the safety standard looks at is not the vehicle alone — ISO 3691-4:2023

Turning to safety, the same structure appears again.

AGVs and AMRs sit inside the same standard

ISO 3691-4:2023 — Industrial trucks, Safety requirements and verification, Part 4, Driverless industrial trucks and their systems — sets out safety requirements and verification methods for driverless industrial trucks and their systems. It is the second edition and has been available since June 2023; the first edition dates from 2020 (global).

Among the terms the standard names as examples of what it applies to are automated guided vehicle (AGV), autonomous mobile robot (AMR), bots, automated guided cart, tunnel tugger and under cart.

In other words, the belief that “an AMR navigates autonomously so a different standard applies” is not correct. Both fall inside the scope of the same standard. Where a vendor proposal contains an explanation along the lines of “this is an AMR so the safety standard applies differently”, it is worth asking specifically what is being referred to.

What the standard explicitly places outside its scope is trucks guided by mechanical means only, and remotely controlled trucks.

The standard looks at the system

What connects directly to the argument of this article is the standard’s definition of a driverless truck system. The standard defines it as being composed of the control system (on board the vehicle and/or off board), the means of guidance, and the power system.

Note that “off board” is inside the control system. What the standard is looking at is not the vehicle in isolation but the whole system, including the master control above it, the infrastructure that provides guidance, and the electrical supply.

The practical implication is considerable. Where a vehicle catalogue states conformity with the standard, that is a statement about the vehicle as a component. It is not a guarantee that the system formed by putting that vehicle into your aisles, on your floor, under your master control and on your power supply meets the safety requirements. How conformity at system level is to be confirmed is something for the buying side and the vendor to settle between them.

The performance levels the standard requires of safety functions are cited specifically for detection of persons, operating modes and braking devices. Of these, detection of persons and braking are closely tied to aisle width, floor friction and running speed. Which means that whether the safety function requirements can be met also depends on the conditions of the layout.

Annex A — preparing the operating area is the buying side’s job

And what gives everything this article has said about routes, aisle widths and floors a footing in the standard is Annex A.

Annex A of the standard holds that the condition of the operating area has a significant effect on the safe running of driverless trucks, and specifies preparation of the operating area in order to eliminate hazards (global).

Put another way, the structure of the standard reflects an allocation in which putting the floor and the aisles into a fit condition is the job of whoever owns the area, not of the vehicle maker. “The floor is poor so we will slow down” is not a vendor dodging responsibility; the structure of the standard makes it the natural consequence.

For the buying side, that translates as follows.

• Defining the conditions of the operating area, measuring them and carrying out the necessary remediation falls inside the buying side’s responsibility
• What to ask of the vendor is the design and verification of safety functions given that area as a premise
• So a proposal that arrives while the conditions of the area are still undefined has a blank where its safety basis should be as well

The claim that layout comes first is about cost and cycle time, and at the same time about assembling the preconditions for safety design.

What you can actually check

Before getting into the detail of the standard, here are things the buying side can verify.

• Which standard and which edition the proposed system claims conformity against
• Whether the conformity claim covers the vehicle alone or the whole system
• For detection of persons, what detection range is set and what the behaviour on detection is (stop, slow, route around)
• Switching between operating modes (automatic, manual, maintenance) and how those privileges are managed
• How it is confirmed that braking distance holds under the assumed floor condition and payload
• What verification is performed after installation, by whom, and how it is recorded

Settling the last item before contract is strongly worth doing. Where post installation verification and its records are left vague, there is no way afterwards to reconstruct what had been confirmed.

Four cost lines that move with the quality of the layout design

The cost discussion here is deliberately confined to the subject of this article. Vehicle price bands and the overall cost structure of a project are covered in how to judge an AGV or AMR project and the order of magnitude of the cost, so what follows looks only at the cost lines whose amounts move with the quality of the layout design.

There are four of them.

What the four have in common

These four share a property. Each of them is a cost line that becomes a line in the quotation if the layout is settled, and an additional claim after the work if it is not.

And of the four, electrical work and safety devices scale with the number of locations, while floor works scale with the area covered. Remove one intersection and one set of safety devices goes with it. Remove one charger and one location of electrical work goes with it. Layout design is, in substance, the work of reducing those location counts.

The assumptions to align when comparing quotations

When quotations are collected from several companies, comparing the totals means nothing on its own. Confirm first that the following assumptions match.

• Which quotation includes floor works. Where it says “by customer”, that cost has to be estimated separately on your own side
• The scope of electrical work. Whose scope covers the primary side wiring up to the charger
• Modification of existing equipment. Lifts, shutters and conveyors are usually work performed by their respective makers
• The number of safety device locations. How many locations the figure covers
• The scope of host integration. What is in this phase and what is deferred
• The assumed utilisation. What percentage sits behind the fleet size

The last item can turn out to be the largest single cause of a difference in price. Assume a high utilisation and the required fleet shrinks and the quotation total falls. It is worth checking that a proposal which looks cheap is not simply built on an unrealistic utilisation assumption. Requiring the basis for the fleet size is also the condition that makes a price comparison possible at all.

Building a payback case for transport automation in Thailand

Anyone looking at transport automation for a site in Thailand runs into the payback discussion sooner or later. There is a structural reason for that.

Labour reduction on its own does not pay back at Thai wage levels

The minimum wage in Thailand stands at 337 to 400 THB per day depending on province, unchanged as of July 2026; the move to a flat 400 THB nationwide has not been implemented. Converted to an hourly figure, that is roughly 50 THB per hour (Thailand-specific; wage levels in Vietnam, Indonesia and the Philippines differ and the figures here do not transfer).

With that as the baseline, an automation investment justified purely on labour reduction will in most cases not pay back as presented. Bringing a return on investment document built on Japanese domestic labour cost into a Thai site breaks down at exactly this point.

Transport is no different. Building a capex request for transport automation on the single argument that transport operators can be reduced usually puts the payback period outside any realistic range.

Running the numbers

Here is the skeleton of the calculation. What follows is an illustrative calculation intended to show the method. It does not represent any particular project. Substitute your own measured values.

Assumptions

• Scope: inter process transport (line replenishment and empty crate collection)
• Current state: three operators’ worth of labour per shift is allocated to transport
• Operating pattern: eight hours per shift, two shifts
• Operating days: 240 per year
• Labour rate: 50 THB per hour (Thailand-specific)

Current transport labour

• Per day: 3 people × 8 hours × 2 shifts = 48 hours per day
• Per year: 48 hours × 240 days = 11,520 hours per year
• In money: 11,520 hours × 50 THB = 576,000 THB per year

Reduction after AGV implementation

Not all of the transport labour disappears. Handling exceptions, preparing loads, assisting transfers, cleaning and inspection all remain. Here the assumption is that 60 percent of the transport labour is replaced.

• Labour reduced: 11,520 hours × 0.6 = 6,912 hours per year
• In money: 6,912 hours × 50 THB = 345,600 THB per year

How to read the payback

Let I be the investment and let R be the annual running cost of the AGV system (electricity, maintenance, the amortised share of battery replacement, master control system support). Simple payback is then

Simple payback in years = I ÷ (345,600 minus R)

Even taking R as zero for the moment, paying back within two years requires the investment to stay within 691,200 THB, and paying back within three years requires it to stay within 1,036,800 THB.

And that investment figure has to cover not only the vehicles but the floor works, the electrical supply for charging, the host integration development and the safety devices at intersections. Since R is never actually zero, the ceiling is lower still.

What this calculation demonstrates is not a prediction of any amount. It is the structural point that a single baseline of labour reduction leaves a very narrow envelope for the investment.

Payback comes from one of three places

So how does an investment decision for transport automation get made in Thailand? The elements that make payback work sit outside labour reduction. The critical discipline is not to add these on top of labour reduction. Adding them together produces an estimate that counts the same benefit twice. Fix on a single baseline and choose which effect is the main axis.

Candidate axis one: increase operating time

Once people no longer have to be assigned to transport, transport can run during periods when people cannot be assigned. Unattended night transport, transport during breaks, transport across a shift change. These are not a reduction in labour cost; they are an increase in the time available to produce, which is a different kind of benefit altogether.

The metric to evaluate here is not hours saved but the additional production time gained and the output that comes with it. Set the benefit side of the estimate, the denominator of the payback period, in terms of production volume from the outset, and do not count labour reduction at all.

Candidate axis two: replicate across processes

Take a configuration established in one process and extend it to a second and a third. From the second deployment onwards, the master control system, the floor remediation already carried out, the electrical installation and part of the host integration can be reused, so the additional investment per vehicle falls.

In that case the investment decision is taken on the whole programme rather than on the first phase alone. One caution applies. If the argument is that later units are cheaper, that has to be confirmed by actually doing the division. Extending into a zone beyond the area already remediated brings new floor works and new electrical work for that zone, and the cost per vehicle may not fall at all. Estimate each destination separately, splitting what can be reused from what cannot.

Candidate axis three: pay back through a different cost line

Transport automation sometimes moves cost lines that have nothing to do with transport. Reduction in work in progress, floor area recovered by clearing temporary set down from aisles, termination of an external warehouse contract, fewer losses from damaged or misrouted goods. These sit in different accounts from labour, and the amounts can be larger.

Where one of these is the main axis, the current value of that cost line has to be measured before implementation. “It probably went down” after the fact is not a basis for a capex approval.

Check the BOI production efficiency improvement measure

As a way of widening the investment envelope, checking the incentive schemes belongs in the plan.

Under the Thailand Board of Investment measures for production efficiency improvement, the corporate income tax exemption cap against the investment amount is 50 percent as the base case, and reaches 100 percent only where the project involves automation or robotics and at least 30 percent of the value of the machinery being replaced is procured from the automation industry within Thailand. The period is three years (Thailand-specific).

The second half of that condition, the domestic procurement ratio, is where projects divide in practice. A configuration that imports the vehicles from overseas and procures only the installation work locally will find the requirement difficult to meet. Where the locally generated portion is larger — building the master control system, floor works, electrical work, installing safety devices — the project moves closer to the requirement.

Which means the layout design changes the composition of domestic procurement, and that composition can affect whether the incentive applies. Because the applicable conditions, scope and application procedure follow the latest guidance from the responsible agency, this is raised here only as an item to check, but it is worth confirming before the investment plan is fixed.

Three common errors in payback calculations

Error one: counting the whole of the labour saved

Transport labour never reaches zero. Exception handling, setup, inspection and cleaning remain. The proportion that is genuinely replaced has to be set from measurement or observation.

Error two: not subtracting the running cost

Electricity, maintenance contracts, battery replacement, master control system support. None of these appear in the initial cost column of a quotation, and all of them recur annually. Calculating a payback period without subtracting them from the denominator produces a shorter figure than reality.

Error three: not deciding where the saved hours go

Even when hours are freed from transport, no financial effect appears unless those people are assigned other work. Absorbing an increase in production without adding headcount, or redeploying to another process. Where that is undecided, the assessment after go live becomes “it is easier, but the numbers have not changed”.

The investment climate and market direction in Thailand and ASEAN (2026)

Finally, a look at the external environment. The figures here are material for understanding what is happening around you, not a basis for your own investment decision.

Investment applications in Thailand

According to Thailand Board of Investment statistics, in the first half of 2026 there were 132 applications under the “Smart and Sustainable Industry” measures, with an investment value of 17.158 billion THB (about 507.6 million USD) (Thailand-specific). The content covers machinery renewal, adoption of digital technology, and the introduction of automation and robotics into production and services.

Total BOI applications over the same period came to 1.47 trillion THB (43.6 billion USD), with approved projects reported to create more than 82,000 jobs for Thai nationals and annual consumption of domestic raw materials of about 386 billion THB (11.4 billion USD) (Thailand-specific).

These figures show application counts and investment values. They do not describe the conditions under which an incentive applies. They are separate from the production efficiency improvement conditions described in the previous section, and the two should not be conflated. What the counts and amounts support is a single reading: capital investment related to automation and robotics continues to be applied for in Thailand.

Market size figures are research house estimates

Several research houses publish market size estimates. All of the following are research house estimates, and the definitions of what each covers differ by company (global).

MarkWide Research estimates the market for AGVs and AMRs in logistics at 8.7 billion USD in 2026 and 34.13 billion USD in 2035, with a CAGR of 16.40 percent.

MarketsandMarkets estimates the market for autonomous mobile robots at 2.75 billion USD in 2026 and 7.07 billion USD in 2032 (the company gives a compound annual growth rate of 14.4 percent, though depending on how the base year is taken those three values do not reconcile with each other; they are figures to read as direction rather than as levels).

These two cannot be placed in the same table and compared. The product scope, the geography and the period all differ. What can be read from them is not a comparison of levels but the single point that multiple research organisations are looking at double digit growth.

In an investment decision it is better not to use market forecasts of this kind as a basis. Your own measured values — transport volume, hours, floor condition, electrical capacity — are far more reliable material for a decision.

What is happening across ASEAN

According to the market reports, more than 20 smart logistics parks are said to have been established in Vietnam and Thailand, with fleets of mobile robots integrated into each facility (regional; this is reported in market research literature and should be treated as secondary information). The direction it indicates, that facilities built around transport automation are increasing across ASEAN, is worth knowing.

The practical implication to hold onto for sites in the region concerns the supply of system integration engineers. As automation projects increase, securing engineers who can design, install and maintain them becomes harder. Pressure on the domestic pool of system integration staff has already been noted in Thailand, and there are situations where the lead time from order to installation is set by the availability of people rather than by vehicle delivery. Taken together with the limited installation windows, this argues for starting the planning early.

Pre order checklist — the order in which it goes onto the drawing

What follows organises the article into the order in which the work is actually done. Most of this can be completed in house before any vendor is contacted. And the more of it is complete, the more accurate the quotations and the more meaningful the comparison.

Step 1 — make the current transport visible

• Write out, process by process, what moves from where to where and how many times a day
• Measure the time per run, and the headcount and hours currently allocated to transport
• Record the condition that triggers transport (accumulation, depletion, time of day, instruction)
• Photograph the current obstructions in the aisles (temporary set down, waiting trolleys, work spilling into the aisle)

The deliverable at this stage is not a working system; it is a set of measured values of the current state. Those measurements become the baseline for the later payback calculation. Current labour hours can only be measured before implementation.

Step 2 — draw candidate routes on the drawing

• Plot the origins and destinations of transport on the drawing
• Provisionally decide which stretches are one way and which are two way
• Count the intersections. Consider whether any can be removed
• Measure the aisle width of each stretch (the effective width including what is actually stored there, not the dimension on the drawing)
• Identify the stretches shared with people, trolleys and forklifts

Step 3 — measure the floor

• Along the candidate routes, record unevenness, steps, joints, gradient and surface material
• Identify locations with concrete defects or peeling floor coating
• Identify zones where water, oil or dust reaches the floor surface
• No pass or fail judgement is made at this point. That comes once the maker specifications for the candidate vehicles are in hand

Step 4 — have the process side decide the stopping and transfer requirements

• For each transfer point, decide the required stopping accuracy and the transfer method
• Allow for the time the transfer takes
• Set out the variation in load presentation
• Decide the behaviour when the receiving side is full or empty

Step 5 — organise the charging and electrical conditions

• List the candidate positions for chargers on the drawing
• Confirm the routing of supply to each position and the headroom in the existing incoming capacity
• Assume a future fleet size and state the capacity that would then be required alongside it

Step 6 — decide the scope of host system integration

• List the counterparts (WMS, MES, production control, lifts, shutters, equipment)
• Separate what is integrated automatically now from what is deferred
• Write the interface specification into the document for the deferred portion as well
• Identify where existing equipment requires modification, and check with the maker of that equipment

Step 7 — assemble the specification and request quotations

Bring all of the above into a single document. In volume, a few drawings and a few pages of text is enough. Then require the following deliverables with the quotation.

Between proposals that all contain these seven, a price comparison has meaning. Proposals that do not may simply be looking cheap.

Frequently asked questions

Who actually does AGV layout design?

Deciding the routes, the floor, the stopping accuracy, the charging positions and the host integration requirements is the buying side’s job. Producing a vehicle configuration that meets those requirements, designing the master control, running the simulation and doing the safety design is the vendor’s job.

Leave that boundary vague and ask a vendor to “propose the layout as well”, and because the vendor does not know how the process really runs, it has no option but to design on assumptions. Where those assumptions do not match reality, the rework arrives at installation.

Annex A of ISO 3691-4:2023 holds that the condition of the operating area has a significant effect on safe running and specifies preparation of the operating area to eliminate hazards, which points to the same allocation: defining the conditions of an area is the role of whoever owns the area.

Realistically it is joint work, with the buying side measuring the current state and defining requirements and the vendor designing on that basis. The projects that start with nothing supplied by the buying side are the ones where the later additional costs are largest.

How many AGVs do we need?

Throughput alone does not settle it. The required fleet is the number of runs per day multiplied by the cycle time per run, divided by the time available for work — and from that available time you have to subtract charging occupancy, waiting at intersections, and stops caused by exceptions.

All three of the subtracted terms are set by the layout: the number and position of chargers, the number of intersections, and the extent to which aisles are shared with people and trolleys. So until the layout is settled, the fleet size calculation cannot be completed.

Nor does throughput rise in proportion to fleet size. Mutual waiting at intersections grows as the fleet grows, so past a certain level, adding vehicles is less and less likely to translate into added throughput. Where that level sits varies with the shape of the aisle network and the master control approach, so there is no general answer.

In practice, require the result of an operating simulation that includes intersections as a deliverable with the quotation. A fleet size back calculated from catalogue specifications is not a basis for anything.

Can an AGV run on our existing floor?

Not judgeable until the maker specification for the candidate vehicle is compared against measured values from your own floor. Flatness requirements vary vehicle by vehicle with wheel diameter and count, the presence of suspension, running speed and payload.

A specialist flooring contractor states that floor level error can be finished to an accuracy of the order of 3 mm to 5 mm (Japan-specific), but that is one contractor’s finish tolerance rather than a standard, and it is not a required value. The reference for judgement is the maker specification.

The point worth noting is that a floor problem shows up as cycle time before it shows up as danger. The same contractor notes that even slight unevenness interferes with running and forces a reduction in speed. A reduction in speed lengthens the cycle time per run and increases the fleet required.

The floor also stops production if it is corrected after go live. Because installation windows at sites in Thailand are limited to periods such as Songkran and the year end, whether floor works are needed is something to settle at the very start of the plan (Thailand-specific).

How far does AGV to WMS integration need to go?

Whether to implement automatic integration in the first phase and whether to write the future integration into the specification are two separate questions.

Running the first phase on manual instructions can be a reasonable judgement in terms of development cost and start up time. The problem comes when automatic integration is added later and the master control system turns out to have no external interface, so modification is required.

The recommended approach is therefore that the first phase does not have to implement it, but the interface specification for the future automatic integration goes into the phase one specification document. State which data is exchanged in which direction, and require the proposal to be premised on expansion not needing modification of the master control system.

The hardest part of integration design is not the communication method but the trigger condition for a transport order. Move when finished goods accumulate, when stock at the next process falls, or at fixed times from the plan? That choice changes the number of runs directly and therefore the fleet size. And because it is how the process is run, only the buying side can decide it.

Does VDA 5050 conformity mean we can mix vehicles from different makers?

In the sense that the communication interface is aligned, the preconditions for mixed operation are in place. VDA 5050 is a communication interface between the master control system and mobile robots, and where it is supported, vehicles from different makers can be handled by a single master control system. Note that it concerns communication between the master control system and the vehicles; it does not standardise integration with WMS or MES. Version 3.0.0 was published in March 2026 (global).

Conformity does not automatically deliver mixed operation, however. Vehicles differ in dimensions, turning radius, acceleration and braking, and safety sensor detection range, so mixing them means designing aisle width and intersections around the most demanding vehicle.

In the specification document, state the version rather than simply writing “supported”. The official site notes that older versions are no longer recommended and will not be developed further. Since mixed operation is impossible if the vehicles do not support it even when the master control does, check both sides. Rather than a compatibility table in a catalogue, the reliable check is evidence of vehicles from different makers actually running under one master control system.

What is included in AGV running costs?

Items that do not appear in the initial cost column of a quotation but recur every year include the following.

• Electricity. Industrial tariffs in Thailand include different unit rates for on peak and off peak periods, so running cost can be reduced to the extent charging can be shifted. It is not worth shifting at the expense of the transport plan (Thailand-specific)
• Battery replacement. Capacity falls with charge and discharge cycles and the achievable distance shortens. Replacement interval, cost, who performs the work, and the supply route and lead time for parts are all worth confirming before ordering. Where stock is not held locally, an import lead time is added on top
• Maintenance contracts. Vehicle inspection, master control system support, software updates
• Floor upkeep. Continued operation wears the floor along the running lines. Where guidance depends on something applied to the floor, maintaining that is included too
• Exception handling labour. Clearing obstructions, returning vehicles to reference position, recovery after rainy season voltage dips (Thailand-specific in the rainy season detail)

When calculating a payback period, subtract these from the annual saving before dividing. Omitting them produces a shorter payback than reality.

Does transport automation pay back in Thailand?

Justified on labour reduction alone, the envelope is very narrow. The minimum wage in Thailand stands at 337 to 400 THB per day by province as of July 2026, roughly 50 THB per hour (Thailand-specific). A return on investment document built on Japanese domestic labour cost cannot be used as it stands because of this single fact.

There are three main axes that make payback work. Increase operating time (transport during periods when people cannot be assigned, unattended night running), replicate across processes (reusing part of the master control, floor, electrical work and host integration from the second process onwards), and pay back through a different cost line (reduced work in progress, recovered floor area, terminating an external warehouse, fewer losses from damage and misrouting).

The critical discipline is not to add these on top of labour reduction. Counting the same benefit under several headings destroys the basis for the capex approval. Fix a single baseline and choose which one is the main axis.

Alongside that, checking the BOI production efficiency improvement measure belongs in the plan. The corporate income tax exemption cap against the investment amount is 50 percent as the base case, reaching 100 percent where the project involves automation or robotics and at least 30 percent of the value of the machinery being replaced is procured from the automation industry within Thailand, for a period of three years. Applicable conditions and procedures follow the latest guidance from the responsible agency, so confirming before the plan is fixed is advisable (Thailand-specific).

How much does doing layout design first actually change the cost?

No general figure can be given, but the cost lines that move can be identified. They are floor works, the electrical supply for charging, host integration development, and safety devices at intersections.

Of those, electrical work and safety devices scale with the number of locations, and floor works with the area covered. Remove one intersection and one set of safety devices goes with it; remove one charger and one location of electrical work goes with it. Layout design is in substance the work of reducing those location counts.

What the four share is the property that a settled layout turns them into a line in the quotation, and an unsettled layout turns them into an additional claim after the work. Floor works in particular carry a cost beyond the money, because doing them after go live means stopping production.

Conclusion

When an AGV project misses on its quotation and its utilisation, the cause is not the vehicle selection. It is that the fleet was counted before the buying side defined the conditions of the place it would run in.

Here are the main points of this article.

• Fleet size is not a linear function of throughput. The naive formula drops three things from the denominator: charging occupancy, waiting at intersections, and stops caused by exceptions. All three are set by the layout. So producing a fleet size without deciding the layout is impossible in principle
• Adding vehicles does not necessarily add throughput in proportion, because mutual waiting at intersections grows with the fleet. There is no general threshold, so require the result of an operating simulation that includes intersections as a deliverable with the quotation
• The order to decide in is routes and occupied width, floor, stopping and transfer point accuracy, charging and battery operation, host system integration. Each later item is a dependent variable of the earlier ones, and reversing the order creates rework
• Aisle width is not set by vehicle width. Read it as occupied width plus safety clearance plus passing allowance. One way against two way is a trade off between travel distance and intersection complexity
• The floor is a cycle time problem before it is a safety problem. Even slight unevenness forces a reduction in speed, and a reduction in speed increases the fleet required. A contractor’s finish tolerance is in the single digit millimetre range (Japan-specific), but the formal required value comes from the AGV maker’s specification, compared against measurements taken along the candidate routes
• ISO 3691-4:2023 covers AGVs and AMRs alike (global). The driverless truck system it defines comprises the control system on board and off board, the means of guidance and the power system. What the standard looks at is the system, not the vehicle alone. Annex A specifies preparation of the operating area, which grounds the position that the floor and the aisles fall inside the buying side’s responsibility
• VDA 5050 version 3.0.0 was published in March 2026 (global). It is a communication interface between the master control system and mobile robots, and where it is supported a single master control system can handle vehicles from different makers. Aligned communication and successful mixed operation are separate things, and mixed operation means designing aisle width and intersections around the most demanding vehicle. Older versions are not recommended and are no longer developed, so state the version in the specification document
• Four cost lines move with the layout design: floor works, the electrical supply for charging, host integration development, and safety devices at intersections. Electrical work and safety devices scale with the number of locations and floor works with the area covered
• In Thailand, labour reduction on its own does not pay back. The minimum wage stands at 337 to 400 THB per day by province, roughly 50 THB per hour (Thailand-specific). Put the main axis on increasing operating time, replicating across processes, or paying back through a different cost line, and fix a single baseline rather than adding them together

On the external side, the first half of 2026 saw 132 applications worth 17.158 billion THB under the Thailand Board of Investment “Smart and Sustainable Industry” measures (Thailand-specific), and research houses are looking at double digit growth (global). Even so, what belongs at the base of an investment decision is not a market forecast but your own measured values: transport volume, labour hours, floor condition and electrical capacity.

One last time. Decide where the vehicles will run before you choose the vehicles. That is the whole of this article.

Drawing candidate routes, measuring the floor and listing the places a charger could stand can all be done in house before any vendor is contacted. In Thailand, TOMAS TECH has worked on the question of how equipment, processes and host systems connect to each other, through building shop floor systems including the PEGASUS production and energy management system. If you only want to talk through design stage questions — whether vehicles can run in aisles of the width you have, how far an existing core system can realistically be connected, whether an installation window can be met — that is entirely fine, and it is no problem at all if you are still at the very beginning of your consideration. We will listen to the situation at your site and lay out the options for how to proceed. You are welcome to get in touch through our contact form.

References

• VDA (German Association of the Automotive Industry), “Information and download: new version of VDA 5050” https://www.vda.de/en/topics/automotive-industry/vda-5050
• ISO, “ISO 3691-4:2023, Industrial trucks — Safety requirements and verification — Part 4: Driverless industrial trucks and their systems” https://www.iso.org/standard/83545.html
• Floor Agent, “Complete guide to AGV running conditions and running environment” https://www.fa-concrete.com/topics/column/agv-info/
• Floor Agent, “Robotisation of logistics warehouses and the millimetres that decide whether it succeeds” https://www.fa-concrete.com/topics/column/robot_yuka/
• Thailand Board of Investment, “Thailand Secures 43.6bn USD 1H 2026 Investment Surge as Big Tech Accelerates Southeast Asia AI Infrastructure Push” https://osos.boi.go.th/EN/news/2430/Thailand-Secures-43-6bn-1H-2026-Investment-Surge-as-Big-Tec/
• The Nation Thailand, “BOI says first-half investment tops B1.47tn as digital and data centre projects pour into Thailand” https://www.nationthailand.com/business/economy/40068948
• MarkWide Research, “AGV & AMR in Logistics Market Size, Share, and Industry Trends Forecast” https://markwideresearch.com/agv-amr-in-logistics-market
• MarketsandMarkets, “Autonomous Mobile Robots (AMR) Market worth USD 7.07 billion by 2032” https://www.marketsandmarkets.com/PressReleases/autonomous-mobile-robots.asp