AMR and AGV Integration

Motionwell Automation integrates AMR and AGV fleets in Singapore, and on a mobile robot project the vehicle is the part you buy rather than the part that decides whether it works. We have been the authorised SIASUN AGV distributor for Southeast Asia since 2017, and we integrate Youibot and MiR AMRs where a fixed route is the wrong answer. Two delivered systems sit behind this page. The first is a fleet of eight laser-guided SIASUN AGF-L15 vehicles rated at 1,500 kg each, handing pallets conveyor to conveyor into an automated storage system, topping up at four charging stations placed on the travel routes, with SIASUN RoboRoute fleet management coordinating multi-vehicle routing, deadlock avoidance and intersection priority, described in the AGV fleet and ASRS integration case study. The second is a QA laboratory programme, re-ordered in four consecutive years, where a collaborative arm rides an AMR and lands samples to plus or minus 0.5 mm through a three-layer positioning stack that corrects plus or minus 50 mm of docking variance. Station equipment is designed, assembled and tested at our Woodlands Link facility, and the company has delivered more than 150 special purpose machines since 2014 under ISO 9001:2015 and bizSAFE Level 3.

Where we stand, said plainly before you read further. We do not manufacture mobile robots or their navigation software, and outside the SIASUN distributorship we are not a robot distributor. We do not supply storage racking or stacker cranes; on the warehouse fleet those came from the storage vendor and we integrated to them. We are not a notified body and do not issue CE certificates. What we design and build is the rest of it, which is where the engineering hours go.

This page covers what integration actually includes, the handoff problem and the ways to close it, traffic and deadlock in shared aisles, charging strategy and what it does to fleet sizing, what the fleet manager has to exchange with your WMS or MES, safety on a shared floor including what changes when a manipulator rides the vehicle, commissioning, and where a fleet is the wrong answer. If you are still choosing between the two vehicle types, that decision is worked through in our AMR versus AGV selection guide and is not repeated here; the wider material-flow scope sits on the warehouse and intralogistics automation page. If you have a station list and a moves-per-hour figure, skip ahead and talk to an engineer.

What Does Integrating a Mobile Robot Actually Cover?

Everything except the vehicle. A mobile robot arrives able to navigate, avoid obstacles and carry a load, and none of that is the project. The project is the set of places it stops, the rules that decide which one it goes to next, the power it draws between jobs, the conversation it holds with the systems that already run your plant, and the parts of the building it has to pass through.

That distinction is worth drawing on paper before anyone quotes, because it also settles who owns each piece.

Read down the third column and the shape of the work is clear enough. One line is a purchase. The rest is design, and it is the reason two quotations for the same vehicle count can be a long way apart. The general form of this scope split, across fixed and mobile robots alike, is on our robot integration services page.

How Do You Hand a Load From a Vehicle Into a Fixture That Wants a Fraction of a Millimetre?

This is the hardest part of a mobile robot project, and it is often left to commissioning. A vehicle navigating by its own map parks within about plus or minus 50 mm. A rack slot, an instrument fixture or a machine nest wants a fraction of a millimetre. Nothing in the vehicle closes that gap, so something else has to.

Three mechanisms close it, and a fourth arrangement avoids needing to.

The laboratory programme uses the first three as a stack, and the layers run in sequence on every pick and place. SLAM navigation takes the base to within plus or minus 50 mm of the station, which is enough to approach and nowhere near enough to hand over. The collaborative arm then works in a coordinate frame calibrated to the docking position and holds plus or minus 0.1 mm repeatability at the tool centre point. Finally a wrist-mounted camera reads fiducials on the tray or fixture, calculates the X, Y and theta offset and feeds a correction to the arm controller, bringing placement to plus or minus 0.5 mm. What that buys is tolerance to drift: rack movement, thermal growth and docking variance are absorbed in software rather than designed out in steel. Each layer retries its own tolerance check up to three times before escalating to an operator alert with the failure mode logged, which is what keeps an unattended overnight run alive through a transient reflection. The full build is in the QA lab automation case study, and the same tolerance stack in a fixed-station context is on our laboratory automation page.

The fourth row is a different move, and on pallet work it is usually the right one. Rather than making the vehicle accurate, make the handoff insensitive to position: the warehouse fleet carries a chain conveyor deck on its top platform and transfers pallets sideways into a stationary chain conveyor, both driven at matched speed, with photoelectric sensors at each end confirming the pallet arrived. No fork tines, no mast, no lifting mechanism on any vehicle in the fleet, and the deck is rated at the same 1,500 kg per pallet as the storage side. The consequence has to be designed in up front rather than discovered: every station the fleet serves needs a mating powered conveyor at matched height, so there is no such thing as a free-standing drop point, and adding one later is a conveyor project rather than a route edit. The gripper end of the same problem, where the load is a tray or a part rather than a pallet, is covered on our end of arm tooling page.

How Do You Keep a Fleet From Deadlocking in a Shared Aisle?

Traffic rules, written before the fleet arrives and tuned after it does. A single vehicle needs none of this. A fleet needs all of it, and the point at which it starts to matter arrives earlier than a two-vehicle pilot suggests.

On the delivered fleet, the RoboRoute server holds a live map of every vehicle position and every planned route. When two vehicles would converge at an intersection it assigns priority and issues hold and proceed commands rather than letting the two on-board obstacle detectors negotiate it between themselves. Routes are recalculated against current traffic density, lanes blocked for maintenance and one-way restrictions, and missions are assigned to the nearest available vehicle by position, battery level and estimated travel time rather than by queue order.

Deadlock is the specific failure this prevents, and it is worth naming precisely because it is not congestion. A junction, an aisle segment or a docking bay is a resource one vehicle occupies at a time. Deadlock is a cycle of those: each vehicle holds a segment another one needs and none can release. Two engineering answers exist. Prevent it, by making lanes one-way, by refusing entry to any segment a vehicle cannot leave, and by acquiring segments in a fixed order. Or detect it and reverse one vehicle out, which needs a vehicle that can reverse safely with a load on it and a rule for choosing which one yields. The prevention route is cheaper to validate and constrains the layout more.

Two site realities complicate this and both belong in the traffic design rather than in commissioning notes. First, the fleet does not control everything in the aisle. People, pallet trucks and forklifts move on their own rules, and staging stock in a lane is a habit no software prevents. Second, recovery decides whether operators trust the system. When a pallet is left in an aisle at two in the morning, somebody has to know within seconds what the fleet did, which vehicle is holding which mission, how the obstruction is cleared, and how the vehicle is returned to service without a supervisor. Those procedures belong in the design, not in the handover pack.

One more number is worth reading correctly. The traffic server on that fleet supports more than 20 vehicles without a software licence change, and eight were deployed. That is deliberate headroom rather than a capacity claim: growth becomes a vehicle purchase instead of a software migration. The real ceiling is the floor, because congestion at junctions and charging bays arrives long before the licence limit does.

Why Does Charging Strategy Change the Fleet Size Calculation?

Because it decides how many hours of the day a vehicle is available to move things, and fleet size follows from that far more than from the specification of the vehicle.

The warehouse fleet uses the middle row. Four charging stations sit on high-traffic travel routes. When a vehicle drops below a configurable threshold, typically 30 percent state of charge, the fleet manager routes it to the nearest free station during a gap between transport missions. Spring-loaded contact plates on the vehicle meet the station’s pads, the charger delivers high-current DC, typically 30 to 50 A, and the vehicle disconnects at its target level, typically 80 percent, or earlier if a high-priority mission is assigned. Battery health data, meaning charge cycles, cell voltage balance and temperature history, is logged so packs are replaced on a schedule rather than on a breakdown.

Where this changes the arithmetic is subtle and it is the reason the strategy belongs in the concept rather than in procurement. Under a dedicated-charging model, availability per vehicle is reduced by the hours each one spends off the floor, so the fleet is sized for the transport demand plus that downtime, and a fleet running two shifts pays for it twice. Opportunity charging returns those hours, but it does not create capacity out of nothing: it converts a vehicle availability problem into a route design problem. Charger placement becomes part of the traffic design, the approach to each station becomes another position the vehicle must reach reliably, and a fleet running consistently near its throughput ceiling will find charging windows competing with transport missions. The honest test at concept stage is whether the duty profile has genuine gaps in it. If it does not, the extra vehicle you avoided buying comes back as a queue at a charger.

What Does the Fleet Manager Have to Exchange With Your WMS or MES?

Less than people expect in volume, and more than people expect in precision. A transport request that says move a pallet from A to B is not enough to run on.

A mission has to carry the source, the destination, the identity of what is being moved, a priority, and the condition under which the vehicle may release its load at the far end. What comes back matters as much: accepted, arrived, transfer complete, and failed with a reason that distinguishes a blocked route from a station that never confirmed. The failure worth designing against is not a vehicle that stops. It is a move that physically succeeded while the record says otherwise, because from that point every downstream decision is made against a wrong inventory position.

On the delivered warehouse fleet, transport requests arrive from the customer’s SAP EWM through standard RFC and BAPI interfaces and are queued for assignment. Every pallet is tracked from pickup to delivery, with the fleet server updating the inventory database at each handoff, so traceability exists without anyone scanning anything by hand, and RFID verification at the dock, the storage entry and the staging area confirms identity at each of those points rather than trusting the route. A web dashboard carries vehicle position, battery state, active missions, queue depth and throughput, and exception handling covers automatic rerouting with manual override from handheld terminals.

Each of those boundaries needs a named owner on both sides of it, and the reason is organisational rather than technical: an interface is where two suppliers can each assume the other owned it. Legacy plant equipment on the station side adds a third variable, and what to do when the machine on the other side of the handoff predates the network is on our legacy machine connectivity page.

What Changes in the Safety Case When the Floor Is Shared, and When the Arm Rides Along?

Start where every machine starts, with an ISO 12100 risk assessment, and note what is different about this one: the hazardous space is not a footprint on a drawing. It travels.

A vehicle brings its own conformity and its own on-board protective sensing from its manufacturer. What the integrator owns is the system around it: where vehicles are allowed to travel, how fast in each zone, what happens where routes cross doorways and pedestrian paths, and how a stopped vehicle is recovered. The QA laboratory shows the layered version of that, and it was a deliberate choice. Fencing the automated stations would have been simpler to justify in the assessment and would have cut the lab in two, so the design used on-board AMR scanning, defined traffic zones, and a cell-level assessment of the collaborative application including the gripper and sample geometry. Technicians work in the same room throughout.

Putting a manipulator on the vehicle changes the problem in a way that is easy to under-scope. The collaborative space moves with the base, so a single assessment at a nominal location does not cover the installation. Four consequences follow:

• The application has to hold at every station the vehicle can dock at. Tens of millimetres of docking variance shift the arm’s envelope relative to the bench and the people standing at it, so the clearances have to be checked where the vehicle actually parks rather than where the layout drawing puts it.
• Travelling and reaching are different states, and the interlock between them is a safety function. The arm has to be stowed and its motion inhibited while the base moves, which means a rated function somebody has to specify, calculate and validate rather than a line of sequence code.
• The payload rides too. Contact limits are a property of the tool and the workpiece, not of the arm, so a sample rack, a tray edge or a gripper finger changes the assessment the same way it would on a fixed cell.
• Stopping is a compound problem. The base can stop for a person while the load on the arm keeps its momentum, so the stopping behaviour that has to be measured belongs to the pair, not to either machine’s datasheet.

The standards behind that work are the ones a fixed robot cell uses: ISO 10218-1 for the robot and ISO 10218-2 for the system and its integration, with each safety function carrying a required performance level under ISO 13849-1, calculated and validated rather than asserted. Singapore does not require CE marking, but the Workplace Safety and Health Act places duties on the occupier and on anyone supplying machinery for use at work, and most multinational manufacturers here specify CE conformity as an internal standard anyway. The guarding, interlock and validation scope that follows is on our machine safety and CE marking page.

What Actually Happens When a Fleet Is Commissioned?

Mostly mapping and rules. The robots arrive working, and the schedule is consumed by everything they have to work with.

Two of those rows are worth expanding, because both can be done in a way that proves nothing. Mapping on an empty floor produces a map that is wrong the moment production restarts, since a warehouse changes shape as stock moves. And an acceptance test that measures one vehicle doing one move measures nothing about a fleet. The run that means something is a sustained one at the real mix, long enough for vehicles to reach their charging threshold, with at least one deliberate obstruction in it to prove the recovery procedure rather than the recovery feature.

Ownership of the map and the traffic rules has to be handed over explicitly, with a named person on your side and a documented way to add a station. A fleet is a system that gets edited, and a fleet nobody on site can edit becomes a fleet that stops being used after the third layout change.

When Is a Mobile Robot Fleet the Wrong Answer?

We deploy these fleets, so read this as the argument against our own scope.

A short fixed link with a steady flow. Where material moves between two points that will not change, at a rate that justifies dedicated equipment, a conveyor is more deterministic, needs no traffic policy and has no battery. A fleet earns its keep on many-to-many flows, not on one-to-one ones.

The move count is low and irregular. Vehicles idle cheaply, but the stations, the interfaces and the safety scope cost the same whether the fleet runs ten moves a day or two hundred. Where the transport demand is genuinely small, the arithmetic rarely closes, and the inputs for testing that are set out in our warehouse automation ROI guide.

The route depends on something nobody controls. An aisle used for staging, a dock that fills with pallets at shift change, a door propped open for ventilation. These are operating habits, and a fleet does not change them; it just stops in front of them.

The floor is not ready. Slopes, gratings, expansion joints, dock plates and unlevel thresholds decide viability before any platform comparison starts, and they are cheaper to find on a site survey than during commissioning.

The destination tolerance cannot be relaxed and the station cannot be changed. If nothing at the receiving end can be modified, and no vision or compliance is permitted in the handoff, the honest answer is fixed automation with the transport handled another way.

Two exclusions while we are being direct. We do not build production welding cells, and we do not sell software products, so the configuration and station logic we write ship with the system. Where a proven standard solution covers the job at a lower price than anything we would build, the useful answer is to say so.

What Should You Settle Before a Fleet Is Quoted?

Seven items, in the order they change the design. The first three size the fleet; the last four are what separate two quotations for the same one.

1. The station list. Every place a load is picked up or set down, what equipment is there today, and at what height its transfer surface sits.
2. Moves per hour by route, at peak rather than on average, with the dwell at each end. Peak is what sizes a fleet; the average is what makes a business case look easy.
3. The heaviest load and how it is carried. A pallet on a deck, a tote, a tray, a rack of samples. The carrying method decides the vehicle top module before anything else does.
4. The tolerance the destination actually needs, and whether the station can be changed instead of the vehicle being made more accurate. Settle it before a layout exists, because it decides which of the four handoff arrangements you are buying.
5. Which system raises a transport request today, what it knows when it raises one, and who is allowed to cancel or override it.
6. Who else is in the aisles, with what equipment, at what hours, and where they stage things when they are busy.
7. What the route crosses. Doors, shutters, lifts, ramps and thresholds, with an owner named for each interface.

If you can answer four of the seven, that is a normal starting point and the rest is what a site survey produces. What we build against those answers is the station equipment, the traffic and charging design, the interfaces to your systems and the safety scope around them, integrated with whichever vehicle the first three answers point at.

Frequently Asked Questions

Do we have to cut our floor to run an AGV?

It depends on what the navigation is trusted to do on its own. On the delivered warehouse fleet, laser guidance was backed by RFID tags grouted into shallow recesses in the concrete at every intersection, docking position, charger approach and dock, because a warehouse changes silhouette as stock moves and wheel-encoder dead reckoning drifts between corrections. The tags sit below floor level so cleaning equipment and remaining forklift traffic cannot damage them. A free-navigating AMR needs none of that work, which is why floor access, curing time and shutdown windows belong in the comparison rather than beside it.

Do we need a WMS before we can run a fleet?

No, and the right answer depends on scale rather than on principle. The warehouse fleet takes its transport requests from the customer's SAP EWM over standard RFC and BAPI interfaces, because the work orders already lived there. The QA laboratory has no MES or WMS at all: the storage logic runs natively in an Allen-Bradley PLC across 70 positions in 7 racks, with a server dispatching missions over Ethernet/IP. The PLC-native route suits that scale and would not suit a site-wide system, and where a validated MES already exists the interface is the answer rather than a rewrite.

Can a mobile robot pass through a fire door or use a goods lift?

Only if something opens the door or calls the lift for it, and that interface is a scope item rather than a vehicle feature. A route that crosses a shutter, a fire door or a floor turns a transport project into a building-services one: the vehicle needs a call and release signal, the door or lift needs an interlock that is safe with an unattended vehicle in it, and the fire strategy for that opening has to survive the change. Establish who owns each of those at the site survey, because the answer decides whether the route is viable before any platform is compared.