Blog

2026.11.09

AGV–WMS Integration in Thailand: Task, Stock and Acceptance Design

AGV–WMS Integration in Thailand: Task, Stock and Acceptance Design

The first question in an AGV–WMS integration project is not how many vehicles to buy or which protocol to use. It is who decides which load must move, from where to where, and what evidence changes a transport request into a completed warehouse movement. Without that decision, a vehicle may reach its destination while WMS stock still sits in the source bin—or the WMS may release stock that has not physically arrived.

This guide is for manufacturing plants and warehouses in Thailand connecting an existing WMS to AGVs or AMRs. It concentrates on the handoff of warehouse tasks, physical load verification, inventory events, exception recovery, RFP requirements and FAT/SAT. It complements our guides to warehouse location management and in-plant logistics improvement. Vehicle selection, fleet interoperability and docking accuracy remain separate design questions.

Define the AGV–WMS integration boundary before choosing an API

A typical architecture has an ERP or MES, a WMS, an integration layer or fleet manager, mobile robots and transfer equipment. ERP may own commercial and accounting stock, MES may own production orders and work-in-process, and WMS may own bins, handling units and warehouse execution. This is a useful starting model, not a rule imposed by product names. Some ERP inventory modules execute warehouse tasks; some MES deployments track work-in-process locations. Record the actual owner of every state and event in the RFP.

The WMS should describe the business move: load ID, material or lot, source, destination, quantity, priority, deadline and task ID. Fleet control translates that request into a vehicle mission while accounting for routes, traffic, charging and equipment availability. The robot reports its own movement and load-handling state. A conveyor, lift or automatic door may have another controller. “Requested,” “accepted,” “picked up,” “arrived,” “handed over,” and “stock confirmed” must remain distinct states.

The VDA’s official VDA 5050 Version 3.0.0, published in March 2026, defines job and status communication between central fleet control and mobile robots. It is not a direct WMS-to-AGV API and does not define when warehouse stock is posted or ERP accounts are updated. Even if a vehicle supports VDA 5050, the WMS-to-fleet adapter, load verification and recovery logic still need design and testing. Ask each supplier which specification version and functions its implementation actually supports.

AGV–WMS Integration in Thailand: Task, Stock and Acceptance Design - figure 1

Do not equate a transport mission with a warehouse task

A WMS task carries business constraints. Can quality-held stock move? Which lot or expiry rule applies? Does the pallet fit the vehicle and destination rack? Is the destination reserved by another order? A successful vehicle trip proves none of these by itself. SAP EWM documentation, for example, treats warehouse task or order confirmation as part of warehouse execution and records confirmation time. That is a product example, not a claim that every WMS uses SAP’s data model.

Where possible, send a business movement request rather than hard-coded route coordinates: pallet P from bin A to line-side station B, with deadline, priority and handling conditions. Fleet control maps these to physical nodes. If WMS stores vendor vehicle IDs or map nodes directly, a map change or fleet replacement may force business-system changes. If direct integration is appropriate, document who owns the mapping, who approves updates and how compatibility is checked.

Start with one defined use case, but test more than “one vehicle moves.” Receiving, replenishment, inter-process moves and finished-goods pickup have different stock posting points and exceptions. Define the chosen use case’s start, end, allowable loads, manual fallback and exception owner. Measure transport frequency and unit load first; some unnecessary movements can be removed through process changes before automation.

Align task IDs, load IDs and location IDs

Keep the WMS task ID, integration request ID, fleet mission ID, robot ID and pallet or container ID as separate fields. They need not have identical values, but every log should be traceable to one business task. If IDs can repeat, include issuer and issue time so yesterday’s job is not mistaken for today’s.

A WMS bin, an equipment transfer point and a fleet map node are different concepts. Maintain a versioned mapping with owner, valid dates, permitted load types, movement direction and equipment conditions. Decide whether a closed destination holds the task or triggers reassignment. Include a test for changed rack or map data in SAT.

Barcode, 2D code or RFID can identify loads. What matters is where it is read and how a missing or wrong read is handled. A source scan can show that the requested pallet exists; a transfer sensor can show that a physical handoff occurred; a destination scan can show that the expected load reached the expected point. For an automatic handoff without a scanner, evaluate load-present signals, weight checks and transfer-point identity. No single signal proves the whole business transaction.

Specify WMS-to-fleet events as a state machine

Write a transition table before implementation. Even a normal move has stages: task created, request accepted, robot assigned, source reached, load received, travelling, destination reached, load discharged, destination verified and WMS task confirmed. Fleet “arrival” is neither equipment “load accepted” nor WMS “stock posted.” Test the three separately.

StageMain ownerEvidenceCondition for next stage
Create and reserveWMSTask, load, source, destination, reservationValid stock and destination
Accept missionFleetRequest and robot assignmentNo duplicate mission
Source handoffEquipment, scanner, robotLoad identity and pickup signalCorrect load on vehicle
Destination handoffRobot and equipmentArrival, unload and load-present signalsEquipment receives load
Confirm stockWMSDestination verification, operator/time/taskBusiness rules satisfied

A WMS may issue stock out at the source and stock in at the destination, or hold a transit state until final confirmation. Either model needs an answer to “where does stranded stock appear?” If transit is a virtual location, define allocation, counting and quality-hold rules. A battery failure must not leave the system saying the load is already in the destination bin.

AGV–WMS Integration in Thailand: Task, Stock and Acceptance Design - figure 2

Reconcile inventory events assuming retries and duplicates

Networks fail and messages are retried. Define an event ID, task ID, event time, receipt time, issuer, schema version, load, source, destination, quantity or handling unit and result. The receiver must recognise an already processed event so that a duplicate completion does not move inventory twice. This allows safe retries without pretending that network delivery is exactly once.

Different IDs can still describe a conflict. If two robots claim to deliver the same pallet for the same task, hold the job for investigation. If a delayed arrival message appears after an operator completed a manual move, do not auto-apply it. Check the current task state, physical location, ordering and cancellation history first. Decide whether an out-of-order event waits, is rejected or requires review.

The GS1 EPCIS/CBV implementation guideline recommends choosing which business steps need visibility events and whether physical items can be identified and read at those steps. EPCIS can supply event vocabulary, but adopting it does not automatically connect WMS and AGV systems. Decide which reads are operationally feasible in this plant. If a label is replaced mid-process, trace the identity conversion.

Before issuing the RFP, trace one real pallet from reservation at the source rack to verified receipt at the line. Treat WMS reservation, fleet acceptance, vehicle arrival, physical loading, destination arrival and receiving scan as separate events. Specify exactly which event releases a reservation, changes the recorded location and closes the warehouse task. Copying the vehicle’s “arrived” state directly into a warehouse “received” state cannot distinguish a blocked transfer point from the wrong load. A fleet completion message may trigger reconciliation, but should not automatically authorize the WMS stock posting.

Maintain a reconciliation record for each task: task ID, load ID, expected source and destination, last physically verified location, last accepted event, hold reason and manual operator. Match the permissions for releasing a WMS reservation to a procedure for finding a held load on site. After a network outage, do not simply trust the newest received message or the event timestamp alone; evaluate the current business state, event ID, cancellation and manual intervention history together. Keep “duplicate ignored” distinct from “awaiting human review” in the audit log.

For example, if a pallet cannot be loaded and is taken to the line manually, include a FAT case in which manual WMS completion conflicts with delayed vehicle loading and arrival messages. Pass only when inventory moves once, the late messages remain visible as held records, and the responsible operator and recovery action can be traced. If a vehicle reaches the destination but no receiving scan occurs, keep transport completion separate from stock confirmation. Attach such cases and expected screen and log evidence to the RFP so vendors quote standard functions, configuration, custom development and site operations on the same basis.

Design exception recovery before go-live

A normal trip is easy to demonstrate. Daily operations depend on what happens when the source is empty, a code cannot be read, the wrong load is picked up, the destination is full, a door stays closed, a forklift blocks a route, communication fails, a vehicle stops or a WMS task is cancelled. For each case, record the physical load location, WMS reservation, mission status, owner and restart condition. “Display an error” is not a recovery procedure.

Separate retry, reassignment, manual transport, return to source and cancellation. Before any action, verify the real load against task and load IDs. When a manual move takes over, stop or invalidate the old mission and define how delayed events are treated. Pressing retry while the original mission may still be running can create two live jobs.

Assign roles: operators verify loads and stop unsafe activity; warehouse supervisors approve stock holds or reassignment; maintenance diagnoses equipment; IT checks integration logs and controlled replay. Define after-hours escalation and conditions for manual supply. Availability targets should reflect the site’s transport rate and loss from stoppage, not a generic percentage copied into an RFP.

Keep equipment and door safety separate from business APIs

WMS integration cannot make a physical transfer reliable on its own. At a conveyor, test load-present, ready-to-receive, permission-to-run, transfer-complete and fault signals. For an automatic door, coordinate robot approach requests, opening permission, closure conditions and emergency egress with the equipment safety design. Measure stopping position, transfer height, load overhang and floor condition at the real site.

A WMS API response must not be treated as proof that a door or conveyor is safe. Local PLC or equipment control handles interlocks, fleet control uses availability, and WMS receives the resulting business event. The OPC Foundation’s OPC UA specification describes a common infrastructure for industrial information exchange across equipment, MES and ERP. It does not by itself define door safety or stock posting. Specify signal meaning and fail states separately.

Where pedestrians and forklifts share routes, assess sight lines, floor level changes, crossings, charging areas, emergency stop and restart. Vehicle and site safety require a project-specific risk assessment by competent specialists. FAT software success does not replace SAT in the real mixed-traffic environment.

Collect baseline data from the Thai plant

Observe the selected route across operating shifts and relevant peaks. Record origin and destination, trips by time, load-size and weight range, empty-container returns, queues, interference and manual exceptions. Two weeks can be a planning example; a site with month-end or seasonal peaks needs representative observations. An average daily trip count can hide a shift-change surge.

Extract WMS task types, create and confirm times, cancellations, reassignment, stock discrepancies and location usage. Walk the floor to find paper or verbal moves absent from WMS logs. The baseline should include production waiting, wrong delivery, count corrections, relief labour and night support—not merely travel time. A faster robot cannot improve output if the destination cannot accept loads.

Confirm terms in Japanese and Thai where both are used. “Received” may mean pickup by the robot or acceptance by the destination operator. “Put away” may mean that the pallet is near a rack or that WMS has posted the bin. Use a shared glossary for screen states and operating procedures.

Put these interface requirements into the RFP

Start with an explicit use case and exclusions: for example, pallet replenishment from a specified raw-material bin to one assembly supply point, excluding held stock, hazardous goods, manual emergency moves and shipping. List the system and version, owner and supplier for WMS, ERP, MES, fleet, robots, PLCs and scanners. A quote that says only “WMS integration” may include a one-way request but omit verification and recovery.

Request a data dictionary for every interface: task and event IDs, load, locations, quantity and unit, priority, deadline, status, timestamps and time zone, error code and schema version. State which fields are mandatory, who issues them, who may change them, what happens when they are missing and how long they are retained. Demand sample messages for normal, duplicate, reordered, cancelled, timed-out and replayed cases.

RFP areaSupplier responseAcceptance evidence
Source of truthState ownership and transition chartSame task traceable across logs
Load and locationReader and wrong/missing read rulesWrong load never posts stock
Retry and idempotencyEvent ID and duplicate handlingDuplicate never posts twice
Cancellation and manual fallbackRules while loaded or travellingLate event held after manual move
EquipmentPLC, door and conveyor signalsNo movement without permission
MaintenanceLogs, backups, access and versioningChanges can be retested

Separate vehicle price from WMS changes, fleet licences, PLC work, scanners, network, charging, equipment modifications, training, service and manual contingency. Ask what a future vehicle addition or WMS upgrade costs. Capacity and ROI estimates should use this site’s load and peak data.

FAT: test messages and failure modes before delivery

FAT can exercise the WMS–fleet connection in a test environment or simulator even if every vehicle and piece of equipment is not present. Freeze software versions, settings, masters, simulator and test data. Rockwell Automation’s official acceptance-testing manual calls for cases and pass criteria tied to requirements, including interface, degraded and exception testing. We use that as a testing principle; its document is for control systems, not a WMS-AGV product specification.

The minimum set includes normal movement, request rejection, retry of the same request, two live requests for one load, wrong scan, cancellation during travel, reordered events, reconnect after timeout, buffering during WMS downtime and unauthorised manual confirmation. For each case record input, starting state, expected WMS stock, fleet and vehicle state, logs and recovery. An error banner alone is not a pass: wrong inventory posting must be prevented and the load must remain traceable.

Reset test data between runs and retain evidence. After a defect is fixed, rerun affected cases. If a supplier says an item depends on the external WMS, assign owners on both sides for a joint test. A screenshot of only one system cannot prove the interface worked.

SAT: accept real loads under site conditions

Use representative labels, pallets, racks, stations, doors, network and operating peaks. Start with one low-risk route; then add concurrent tasks, priorities, blocked paths, equipment stops, network loss, low battery and manual fallback. Capture the physical load location, WMS stock and reservation, fleet mission and equipment signals together. Vehicle arrival time alone misses unloading and WMS confirmation delay.

Set site-specific pass criteria: no wrong-load posting, no double posting after reconnection, a stopped load can be located within the agreed time, and the approved fallback keeps supply running. Measure request-to-confirmation time distribution, peak queue, charging, utilisation and manual intervention. Agree on samples, shifts and exclusions before testing.

Safety tests require the equipment and vehicle owners and an approved procedure. After an emergency stop or protective scanner intervention, also check that recovery does not create a duplicate mission or stock posting. Reassess and retest affected cases after map, bin or station changes. Handover documents should include versions, backups, transition chart, results, open issues, contacts and manual recovery instructions.

AGV–WMS Integration in Thailand: Task, Stock and Acceptance Design - figure 3

Operate with end-to-end measures and change control

Separate request-to-accept, accept-to-pickup, pickup-to-unload and unload-to-WMS-confirmation times. This shows whether a delay comes from vehicle shortage, source preparation, transfer congestion or stock confirmation. Track duplicate events, unverified loads, manual recovery and long-held reservations. Review specific task IDs and physical records, not just averages.

A route addition, bin change, new load type, WMS upgrade, extra robot or VDA 5050 version change has different effects. Link each change request to the mapping, data dictionary, APIs, equipment signals, training and tests. The name “VDA 5050 compatible” alone does not guarantee two actual products implement the same version and features; keep real integration-test results.

Set log access and retention with the company’s information policy. Robot position and task logs may be used for fault analysis, inventory audit and workforce management. Decide whether they leave the plant or cross borders. Keep a minimum way to locate a load during outages and reconcile differences after recovery.

FAQ: AGV–WMS integration

Can we deploy AGVs without a WMS?

Yes, a fixed-route mission can start from fleet software or a local call button. The unresolved question is how load identity, origin, destination and receipt affect stock records. Define location rules and retain task and event IDs so a later connection is feasible.

Does VDA 5050 support remove the need for WMS integration?

No. VDA 5050 covers communication between fleet control and mobile robots. Turning WMS tasks into missions, verifying loads, posting stock and handling cancellations remain separate. Verify version, implemented features and the real use case.

Should stock update when the AGV reaches its destination?

Not automatically. Arrival shows vehicle position, not physical discharge or bin confirmation. Define the necessary evidence—unload signal, destination scan, equipment receipt or operator check—and let WMS confirm according to business rules. Define transit stock if needed.

How should outages and duplicate messages be handled?

Store task and unique event IDs and make receiving transactions idempotent. On reconnect, reconcile open tasks with physical loads before applying late events. Do not issue another mission while the original may still be running.

What is the FAT/SAT split?

FAT checks interface logic, state transitions, duplicates, reordering and cancellation in a controlled environment. SAT checks real loads, routes, stations, doors, shared traffic, site network and manual recovery. Both need correlated WMS, fleet, vehicle and equipment evidence.

Conclusion: design from the stock confirmation point

AGV–WMS integration succeeds when warehouse tasks and real loads advance consistently. Define WMS tasks, fleet missions, vehicle movement, equipment handoff and stock confirmation as separate states. Correlate them by task and load IDs, design retry and recovery, test logic at FAT and physical operations at SAT.

If your Thai plant is defining the connection to an existing WMS, start with one route, its task records, location list and exception history. TOMAS TECH can help organise the business events and equipment handoff conditions into an RFP and acceptance plan. Contact us while the scope is still being discussed.

Primary sources