A successful automated guided vehicle implementation in a Thailand or ASEAN factory begins before anyone decides how many AGVs to buy. The buying team must first define what has to move, by when, and what observable state means that the transport is complete. A smooth vehicle demonstration proves little about peak congestion, forklift crossings, charging, WMS/MES mismatches or recovery from faults. This guide gives factory owners, operations, logistics, EHS, IT and procurement teams one practical evidence chain from RFP and PoC through FAT, SAT and operational acceptance.
The executive decision: procure a transport service, not a vehicle
The first controlled object should not be a model or navigation technology. Define the service from the creation of a move request to confirmed load handover. At minimum, specify source, destination, load, weight, required time, priority, transfer conditions, completion signal and the safe return state after an exception. This “transport contract” makes it possible to compare magnetic-path AGVs, laser-guided vehicles, SLAM-based AMRs, tuggers and autonomous forklifts against the same demand.
An RFP that says only “500 kg payload, 1.5 m/s maximum speed, five units” invites incomparable proposals. One supplier may exclude floor repair, another may include the WMS interface, and a third may treat night commissioning as reimbursable work. The price difference then represents undefined scope rather than better performance.
Use five evidence gates.
| Gate | Decision | Minimum evidence |
|---|---|---|
| Before RFP | Is this move suitable for automation? | Current from-to data, volume, waiting, near misses and constraints |
| PoC start | Is the uncertainty worth testing? | Route, success and stop conditions, responsibility boundary |
| FAT | Was the system built to the approved design? | Vehicle, software, I/O and fault-scenario test records |
| SAT | Does it meet the requirement in the real factory? | Records including floor, wireless, people, equipment and peaks |
| Operational acceptance | Can the factory operate it safely without exceptional vendor dependence? | SOPs, training, maintenance, drills, KPI and change control |
Screen the use case before copying an AGV implementation case study
Moves that are often suitable
Repetitive moves with defined sources and destinations, stable loads and measurable delay consequences are good candidates. Examples include scheduled delivery from a parts supermarket to assembly, inter-process finished-goods transfer, empty-container return and pallet movement to inspection. The baseline should capture not only walking time but also machine waiting caused by late material and the time lost waiting for a forklift.
Moves that need process improvement first
Standardise before automating when the destination changes informally, loads frequently collapse, aisles are routinely used as storage, transfer heights vary or operators request transport verbally. An AGV does not correct an ambiguous operation. It reproduces the ambiguity at machine speed and turns it into stop codes.
“A person can move it” is not a sufficient automation requirement. A person straightens a tilted box, moves a cart out of the aisle, notices a wrong label and calls the receiving process. Allocate each of those hidden tasks to vehicle or station functions, an upstream system, or controlled standard work. That allocation defines the real scope.
Build the RFP baseline with ten operating days of measurement
This article proposes observing at least ten working days that include normal and peak conditions. Ten days is a planning recommendation, not a standard. A seasonal factory should also reconcile the observations against historical production and shipment data.
Fields for a controlled from-to table
| Field | Example | Use in the RFP |
|---|---|---|
| Source and destination | ST-A01 → LN-B04 | Map, station and route scope |
| Load | 600 × 400 returnable box, maximum 80 kg | Top module, transfer, centre of gravity, detection |
| Trigger | Empty kanban, schedule or MES instruction | Request interface |
| Demand | Normal 35, P95 52 missions/hour | Fleet and congestion design |
| Time constraint | Within eight minutes of request | Priority and acceptance KPI |
| Current time | Separate travel, queue and handling | Benefit and bottleneck model |
| Exception | No load, duplicate request, closed aisle | Recovery design |
Collect a 15- or 30-minute distribution, not only a daily average. A site may record 100 moves per day while 40 arrive immediately after shift start. That concentration determines vehicles and buffer space. P95 means 95% of observations are at or below the stated value; it is not a mandatory standard. It is used here as a practical way to prevent an average from hiding peaks.
Agree on one definition of completion
“The vehicle arrived,” “the load transferred,” “the PLC returned receive-complete,” and “WMS inventory posted” are different business states. State both physical and business completion in the RFP. Give every mission a unique ID and define idempotent retry so that a message retransmission does not post the same inventory move twice.
AGV layout design means designing the operating zone, not drawing a route
Aisles and intersections
A shortest-path line becomes slow at blind corners, narrow aisles, crossings and passing conflicts. Overlay pedestrians, manual carts, forklifts, tuggers and existing conveyors by time of day. Define one-way segments, priorities, stop lines, crossings, exclusions and waiting pockets. Emergency exits and firefighting access must remain clear.
There is no responsible universal aisle-width number. Determine it from the vehicle envelope, load overhang, swept path, braking and detection fields, human escape possibility, speed, racks and walls, the relevant standards and supplier restrictions. Keep the assumption and the authority to change it with every layout value.
Floor, gradient and environment
Cracks, joints, steel plates, oil, water, dust, reflections, strong sunlight, temperature and humidity can affect motion, stopping, sensing and tyre life. Include rainy-season entrances, post-cleaning conditions, open shutters and maximum loads in validation. The RFP should allocate floor repair between the vehicle supplier, civil contractor and factory.
Transfer stations
Arrival is not completion if the load cannot be handed over reliably. Define height, permissible position error, load-present and station-free signals, stops, doors, conveyor Ready/Busy and timeout at every station. Derive the stopping tolerance from the top module and actual load interface; do not copy a catalogue headline value.
Charging and shifts
Do not accept a single statement such as “eight hours continuous running.” Model mission rate, waiting, acceleration, load, floor, ambient conditions, charger count, charging queues and battery ageing. Verify that charging does not block a route, that power recovery starts from a safe state, and that ownership covers the storage, handling and disposal of exchange batteries when used.
For a broader equipment comparison, read the material-handling equipment selection guide. For a cost breakdown, see the five-year AGV and AMR TCO guide.
Size the fleet from cycle-time distributions and contention, not average throughput
The basic arithmetic is simple:
Required busy minutes/hour = missions/hour × effective cycle minutes/mission
Theoretical vehicles = required busy minutes/hour ÷ 60
The effective cycle must include empty travel, loaded travel, transfer, crossing queues, door or elevator waits and dispatch waits. Charging, planned maintenance, fault recovery and peak reserve are evaluated separately.
An explicitly hypothetical calculation
The following values are assumptions for explanation, not performance evidence, market data or a quotation. If P95 demand is 36 missions/hour and the effective cycle is 6.5 minutes, demand requires 234 vehicle-minutes/hour, or 3.9 theoretical vehicles. Four leaves almost no margin. A five-vehicle option should be simulated and tested in the PoC against charging and variation.
Inspect the slowest five percent of intersection waits instead of relying on the mean cycle. Adding vehicles can reduce output per vehicle because congestion increases; “four were insufficient, so buy five” is not linear. A layout change, priority rule, buffer or levelled release may cost less.
AGV cost must be compared as five-year TCO, not vehicle price
Normalise proposals into one cost structure. Initial cost includes vehicles, top modules, charging, Fleet Control, wireless, floor/electrical/safety work, host interfaces, testing, training and spares. Operating cost includes maintenance, software, batteries, wheels and sensor-window consumables, connectivity, backup, on-site response and regression testing after updates.
| Cost element | Hypothetical THB | Note |
|---|---|---|
| Three vehicles and top modules | 3,600,000 | Illustrative assumption |
| Fleet Control and charging | 1,250,000 | State licence terms separately |
| Station, WMS/MES and PLC interfaces | 1,500,000 | State design matters more than I/O count |
| Floor, electrical, safety and wireless | 900,000 | Depends on site survey |
| FAT/SAT, training and spares | 1,350,000 | Check night work, travel and interpretation |
| Total initial cost | 8,600,000 | Not a market price |
With an assumed annual operating cost of THB 1,250,000, nominal five-year TCO is 8,600,000 + 1,250,000 × 5 = THB 14,850,000. Tax, financing, inflation, residual value and production-loss risk are excluded. This is a transparent calculation example, not price guidance.
Treat benefits with the same discipline. Released walking hours do not automatically become cash labour savings. Separate increased output, avoided recruitment, skilled operators returned to machines, reduced material-wait downtime and risk reduction, and assign an owner for benefit realisation.
Do not include a BOI incentive in the base case before approval. Confirm activity, equipment, applicant, timing and conditions with BOI and competent tax/legal advisers, then show it as a supported case. BOI reported 82 machinery-and-automation applications worth THB 13.1 billion and 132 Smart and Sustainable Industry applications worth THB 17.2 billion in 1H 2026. Those figures provide investment context; they do not establish the eligibility of a particular AGV project.
AGV safety measures are an evidence system, not a scanner purchase
ISO 3691-4:2023 addresses safety requirements and verification for driverless industrial trucks and their systems and explicitly includes AGVs and AMRs as examples. ISO 12100:2010 provides the general risk-assessment and risk-reduction method. ISO 13849-1:2023 covers the design and integration methodology for safety-related parts of control systems. Listing those references in a proposal is not equivalent to making the operating zone safe.
ISO’s current catalogue lists ISO 3691-4:2023 as the published edition while a new DIS is under development. Do not treat a draft as a current mandatory provision. The RFP should state the contractual edition, how future differences will be assessed and who owns migration work.
VDA 5050 3.0.0 is useful for communication between mobile robots and Fleet Control. Its published scope explicitly excludes safety requirements and traffic-management logic. “VDA 5050 compatible” must not substitute for a safety case or a congestion-control test.
Include every operating mode in the risk assessment
- Automatic travel, manual mode, teaching, maintenance, cleaning, towing and rescue
- Maximum, offset and overhanging load; empty travel; load shift and detection failure
- Crossings with pedestrians, forklifts and carts; blind corners, passing and wrong-way entry
- Wireless loss, Fleet Control outage, PLC state mismatch, power loss and restart
- Obstructed or contaminated safety scanner, emergency stop, protective stop and restart permission
- Fire, evacuation and special modes of shutters, doors and elevators
Derive required safety performance from actual risk assessment and the relevant standards. Neither “make every function PL d” nor “use the manufacturer’s standard setting” is automatically correct. Use a Safety Requirement Specification that traces hazard, hazardous situation, protective measure, safety function, verification, residual risk and operational control on one line.
In Thailand, the plant’s EHS function and competent professionals should determine the applicable occupational-safety, machinery/forklift, building, fire and electrical requirements. Do not claim that ISO conformity or a CE mark proves compliance with every Thai requirement. The EU Machinery Regulation 2023/1230, applicable from 20 January 2027, is relevant to EU market access and supplier roadmaps but is not a replacement for Thai law.
Put twelve controlled packages in the AGV RFP
1. Business requirement
State products, volume, time window, shifts, growth, tolerated downtime and benefit KPI. Replace “save labour” with the exact work and shift whose hours will be released and the destination of those hours.
2. Loads and top modules
Give minimum and maximum dimensions and mass, centre of gravity, base, allowable acceleration, load stability, identification and rejection rules. Provide representative loads for FAT.
3. Layout and environment
Provide CAD, coordinates, aisles, gradient, joints, floor, temperature, humidity, dust, water, light, indoor/outdoor limits, work windows and restricted areas. Allocate responsibility when drawings differ from reality.
4. Capacity and fleet
Provide time-bucket demand, peaks, mission priority and maximum waiting time. Make simulation inputs, assumptions and outputs contractual deliverables.
5. Stations and peripheral equipment
Define states, I/O and mechanical/electrical ownership for conveyors, doors, lifts, PLCs and indicators.
6. Host systems
Define WMS/MES/ERP requests, status, cancellation, priority, retry, history, clock and IDs. An API endpoint is not a complete interface; specify outage behaviour.
7. Safety
Make the decision ownership for standards and law, risk assessment, protective measures, verification, residual risk, signs and training explicit.
8. Non-functional requirements
Define availability, recovery, backup, log retention, access, language, time synchronisation, monitoring and network dependence.
9. Cybersecurity
Define accounts, least privilege, remote service, MFA, signed updates, vulnerability notification, support end date, configuration backup and log export. Document the boundary between cybersecurity and safety functions.
10. FAT and SAT
Agree test case, data, acceptance, retest, witness and evidence format before contract. Acceptance criteria invented after delivery become commercial disputes.
11. Training and maintenance
Separate operator, supervisor, first-response, maintenance and administrator roles. Require Thai-language material where appropriate, local spares, response time, night/holiday support and an end-of-product plan.
12. Change control and intellectual property
Decide ownership and export rights for maps, PLC logic, API configuration, reports and configuration/source artefacts. Define price, approval and regression testing for layout changes and added vehicles.
Use a twelve-week PoC to retire uncertainty, not to stage a demo
The PoC must reduce uncertainty that could change the investment decision. It should not merely prove that a vehicle can travel. The following is a proposed programme, not a duration required by a standard.
| Period | Work | Exit criterion |
|---|---|---|
| Weeks 1–2 | Baseline, route, load, hazard and I/O confirmation | Approve inputs and open decisions |
| Weeks 3–4 | Detailed design, map, state transitions and tests | Pass Design Review |
| Weeks 5–6 | Single-vehicle travel, stations, charging and base interface | Repeat normal flow |
| Weeks 7–8 | Peak, multi-vehicle, crossings, priority and congestion | Evaluate capacity hypothesis |
| Weeks 9–10 | Communications loss, failed vehicle, no-load and restart | Prove safe stop and controlled recovery |
| Weeks 11–12 | Real operation, training, KPI and investment decision | Record Go, Conditional Go or No-Go |
List every difference when the PoC vehicle is not the production model, the aisle is specially cleared, an expert supplier engineer remains on site, or the host interface is replaced by a button. Do not extrapolate a PoC result to production capacity while those differences remain unbounded.
Manage FAT, SAT and operational acceptance in one test matrix
Factory Acceptance Test
At FAT, verify the vehicle, top module, Fleet Control, charging, I/O and approved software build in the supplier’s factory or a simulation environment. Test minimum and maximum loads, sensor obstruction, emergency and protective stops, loss of communication, restart, duplicate requests, cancellation and low battery. The real factory’s floor and wireless conditions cannot be fully reproduced, so FAT approval is not site-capacity approval.
Site Acceptance Test
SAT uses the actual floor, routes, people, forklifts, wireless, stations, WMS/MES and shift conditions. Include peak release, breaks and shift change, no-load, full destination, door timeout, one failed vehicle, one failed charger, communication loss and power recovery.
| Test | Example measure | Example decision rule |
|---|---|---|
| Capacity | Valid missions completed | Meet agreed P95 demand for the continuous test window |
| Response | Request-to-arrival time | Agreed percentage within time constraint |
| Completion | Valid missions without unplanned human intervention | Proposed example ≥99.0%; freeze exclusions first |
| Availability | Uptime during agreed production window | Proposed example ≥98.0%; define stop categories |
| Transfer | Damage, drops and wrong delivery | Zero major events; cap minor events |
| Recovery | MTTR and procedural deviation | Trained factory role recovers within SOP |
| Safety | Approved safety-function scenarios | All required cases pass |
The 99.0% and 98.0% values are proposed contractual examples, not standard values. Define the denominator, observation window and treatment of planned stops, upstream equipment outages and invalid loads. A serious safety event is not averaged out by good throughput; it triggers stop, correction and revalidation.
Operational acceptance
Final acceptance also proves organisational capability. A shift supervisor can classify a stop, first responders can handle a failed vehicle safely, maintenance can restore a backup, IT can disable an account and EHS can re-assess a change. Control the language and revision of every delivered SOP, drawing, risk record, configuration and training record.
Evaluate an automated guided vehicle implementation case with seven questions
Another factory’s case study becomes useful only when its operating conditions are visible:
- What moved, between which points, how many missions, and when was the peak?
- How did the system coexist with people, forklifts, doors and elevators?
- What floor, load or standard-work changes were made before launch?
- What were the WMS/MES/PLC boundaries and outage behaviours?
- Which conditions changed between PoC and production?
- What were the three largest stop causes after launch, and what changed?
- Was value realised as released hours, output, avoided hiring, downtime reduction or risk reduction?
IFR’s World Robotics 2025 release reports 102,900 professional service robots sold globally for transportation and logistics applications in 2024, up 14%, and RaaS fleet growth of 42%. These are global professional-service-robot figures, not Thailand AGV sales. Use them only as market context for checking support capacity, spares, data portability and long-term compatibility—not as proof that one factory’s project will succeed.
Failure patterns that prevent acceptance
Treating a one-vehicle demo as fleet capacity
Single travel does not exercise intersection contention, charging, failure or priority. Test a multi-vehicle peak scenario.
Reducing AGV safety measures to an onboard scanner
Load overhang, transfer, rescue, doors, forklifts and human behaviour remain outside a vehicle-only certificate. Assess the whole operating zone.
Calling an API connection a completed business interface
One normal request/response does not prove cancellation, duplication, timeout, retry, restart or clock drift. Test the full state model.
Selecting the lowest vehicle price and adding the system later
When top modules, floor, wireless, charging, station integration, training, night work and maintenance are excluded, vehicle price is not TCO. Normalise all proposals.
Depending on the vendor engineer for every recovery
A night-shift stop that waits for an overseas engineer will miss the availability target. Make factory first response, logs, escalation and spares acceptance deliverables.
FAQ about automated guided vehicle implementation
How much does an AGV implementation cost?
Vehicle price is insufficient. Compare five-year TCO including top modules, charging, Fleet Control, floor, electrical, wireless, safety measures, WMS/MES/PLC interfaces, FAT/SAT, training, spares, maintenance and updates. The THB 14.85 million example in this guide is hypothetical, not a quotation or market range.
What should an AGV layout design decide first?
Set sources, destinations, time-bucket demand, loads, completion states, crossings and transfer equipment before drawing a route. Then design travel, waiting, charging, stopping and recovery as one operating zone.
Is ISO 3691-4 enough for AGV safety measures?
No. ISO 3691-4:2023 is a central Type-C standard, but the project also needs risk assessment based on ISO 12100, appropriate safety-control design, supplier requirements, Thai legal/EHS review, and evaluation of peripheral equipment and the building. Competent parties must assess the actual vehicle and site.
How long should an AGV PoC run?
No universal standard duration applies. This guide proposes twelve weeks to include baseline, design, multi-vehicle peaks, fault recovery and real operation. Good existing data and test facilities may shorten it; complex integration or construction may extend it.
What is the difference between FAT and SAT?
FAT confirms the approved design, functions and fault handling before shipment in a controlled environment. SAT confirms requirements on the actual floor and wireless network with real traffic, peripheral equipment and host systems. Every site condition that FAT cannot represent should be explicitly transferred to SAT.
Does VDA 5050 make mixed fleets plug-and-play?
No. VDA 5050 3.0.0 provides a common communication foundation, but does not define safety or traffic-management logic. Test supported features and versions, optional fields, maps, charging, errors, top modules and support ownership.
Conclusion: preserve the evidence chain from RFP to operations
An automated guided vehicle implementation should procure a transport service that delivers the right load on time and safely, and that can recover from exceptions—not merely vehicles. Establish a ten-day baseline, design layout, transfer, host integration and safety together, and compare five-year TCO. Put responsibility and testing in the RFP, use the PoC to retire material uncertainty, and trace the same requirement IDs through FAT, SAT and operational acceptance. That is how a factory closes the gap between a successful demonstration and sustainable production.
TOMAS TECH can support Thai factories before a vehicle is selected, including baseline measurement, AGV/AMR comparison, RFP and layout design, WMS/MES/PLC interfaces, and FAT/SAT planning. You can contact us while your team is still aligning assumptions for a budgetary estimate.
Primary and official sources
- ISO 3691-4:2023 — Driverless industrial trucks and their systems
- ISO industrial-truck catalogue — published and developing standards
- ISO 12100:2010 — Risk assessment and risk reduction
- ISO 13849-1:2023 — Safety-related parts of control systems
- VDA — Release of VDA 5050 Version 3.0
- VDA 5050 Version 3.0.0 specification
- EUR-Lex — Regulation (EU) 2023/1230
- Thailand BOI/OSOS — 1H 2026 investment applications
- IFR — World Robotics 2025 service-robot release
- ILO NATLEX — Thailand OSH Act B.E. 2554
- ILO NATLEX — Thailand machinery, crane and boiler ministerial regulation
