Under-Rider AGV Selection: Low-Profile Design and FAT/SAT

When a Thailand factory evaluates an under-rider AGV, filtering candidates by nominal payload and vehicle height is not enough. A robot may fit beneath a cart yet fail to lift it reliably, lose traction when the load center shifts, or leave operators without a safe recovery method after a stop. The purchase is therefore not just a vehicle. It is a material-handling system comprising the cart, load, floor, route, operating zone, safety functions, dispatch logic and support process. This guide turns those interfaces into procurement requirements and evidence for FAT, SAT and a clearly hypothetical 90-day proof of concept.

What an under-rider AGV actually automates

A low-profile under-rider automated guided vehicle travels beneath a cart or shelf and couples by lift, pin, hook or another engineered interface. It can reduce manual transfer because the cart remains the load carrier. Some solutions also allow one robot to serve multiple carts; the MiR600 Shelf Lift page, for example, states that one robot can handle multiple carts. That is a concept, not proof that a particular cart, route or throughput will work.

“Fits underneath” does not mean “compatible.” Swivel casters may sweep into the vehicle envelope. Crossmembers may miss the lift point. A repaired cart may have less clearance than its drawing. Off-center loads change wheel reaction and braking behavior. The Meidensha low-platform page illustrates distinct envelopes—one model listed at 180 mm body height and 380 mm body width, and another at 132 mm body height with an 800 kg permitted load—and notes that existing carts do not always need modification depending on the model. It does not promise that every existing cart can be reused unchanged.

This article is deliberately narrower than an AGV basics guide, tugger comparison, roll-container overview or AMR case study. For the broader investment model, see AGV and AMR costs for Thailand factories. Use our AGV vendor-selection guide for sourcing governance and AGV layout design guide for aisle, crossing and charging decisions.

Break low-profile AGV selection into fit, carry, travel and recover

Instead of one “payload” row, structure the decision around four conditions:

1. Fit: vehicle, sensors, antenna and lift enter and exit without interference.
2. Carry or couple: the interface remains stable under centered, offset and dynamic loads.
3. Travel: the coupled system works on the actual floor, slope, joints, doors and mixed traffic.
4. Recover: people can safely manage misalignment, network loss, low battery, wrong-cart detection and disabled vehicles.

Give each condition a requirement ID. Reuse that ID in the RFP, design review, demo, FAT, SAT and handover. Traceability prevents a persuasive demonstration from becoming an untested acceptance assumption.

Start cart automation with a physical cart census

Two carts made to the same drawing may differ after years of welding, caster replacement, bent plates and field modifications. Survey every PoC cart, or define a sampling method that can be defended. Create a cart passport containing ID, photographs, dimensions, empty and maximum mass, load envelope, caster specification, repair history and disqualifying defects.

Measure the complete under-cart envelope

Static minimum clearance is only the start. Measure the swept volume while the AGV turns into the cart, the 360-degree caster sweep, bolt heads, brake pedals, braces, hanging straps and displaced wrapping. Model the worst combination of floor unevenness, cart deflection and dimensional tolerance. A few millimeters added to nominal height is not an engineering tolerance stack.

Sharp’s published TYPE S example is 400 mm wide and 185 mm high, states a 200 kg permissible/under-cart towing load, uses magnetic guidance, lists a maximum 40 m/min and refers to carts at least 850 mm wide. Meidensha publishes an ultra-low-floor lift AGV example at 132 mm high with an 800 kg permitted load. These figures show how different the mechanical envelopes can be. They are not a normalized ranking: load center, test method, cart structure, dynamic duty and route remain different.

Verify load center and cart strength

Total mass below a rating does not establish fitness. An offset load changes lift reactions, drive-wheel contact and braking attitude. Require evidence for:

• empty, normal, maximum and credible offset-load centers;
• lift-pad or pin-receiver material, thickness, reinforcement and deflection;
• acceleration, deceleration, cornering and emergency-stop load restraint;
• caster diameter, material, swivel resistance, brake state and wear limits;
• load height, overhang, visibility obstruction and falling-object control.

Put the boundary among cart supplier, AGV integrator and plant user on an approved interface drawing. That avoids post-acceptance disputes about whether a cart was “weak” or “misused.”

Treat docking and lifting as one sequence

Coupling success depends on AGV XY and angular error, cart parking error, caster direction, slope and lift stroke. Specify allowable and rejectable conditions, pin insertion or lift overlap, coupled confirmation, cart-present detection, wrong-cart rejection and release confirmation. The system must inhibit departure when the sequence is incomplete.

OMRON’s OL-450S public page gives an example of an integrated low-clearance lift with a maximum 450 kg payload and a 108–308 mm lift range. It states 11 hours of runtime unloaded, seven hours loaded, and 25 minutes for 20–80% charging. These are useful design inputs but not a site-runtime promise; route, acceleration, waiting, temperature, battery age and charging policy all matter. The speed value displayed on that page appears anomalous, so this article does not use it.

Roll-cage and trolley automation needs a reject standard

Roll cages often carry doors, shelves, covers, tags, wrap and damaged casters that can intrude into the AGV path. A load can also overhang and strike equipment before the vehicle’s protective field reaches it. Define not only the acceptable trolley but also the trolley that must be rejected.

Parking positions may need wheel guides, hard stops, orientation marks and cart IDs. Maintain empty/full and quality-hold states consistently between physical labels and the control system. If a cart is returned diagonally, the robot should identify the condition, stop safely and provide an actionable message in the local operating language instead of forcing engagement.

Treat the floor and route as part of the low-profile AGV

Low clearance makes small steps, drain channels, lifted plates and debris more consequential. Survey the worst point along the full route, not the average floor.

Record steps, gaps, longitudinal and cross slopes, friction, water, oil, dust, drains, expansion joints, cable covers, lifts and dock plates. Include minimum aisle width, swept turns, escape areas, blind intersections, fire doors, shutters, PLC interlocks, emergency exits, network coverage and localization features. Establish a change-control process for floor repairs, cleaning and layout changes.

Check stopping behavior while coupled to the real loaded cart. Width calculations must include cart and load overhang, localization error, safety margin and space for a person to avoid the system—not merely robot body width. Recovery space is also a design requirement: a disabled loaded cart should not permanently block fire egress or the only production aisle.

Safety is determined by the operating zone, not the vehicle alone

ISO 3691-4:2023 is the published standard for driverless industrial trucks and their systems, including under-cart types. Its public record makes operating-zone conditions material to safety. Buying a vehicle with a safety scanner therefore does not demonstrate that the installed system is safe. The assessment must cover people, forklifts, loads, doors, machinery, floors, maintenance and operating rules.

ISO marks the 2023 edition as due to be revised. A third-edition ISO/DIS 3691-4 is in draft and adds or expands matters including side detection, commissioning content and warnings. A DIS is not a final standard and may change. An RFP should ask the supplier to identify the edition applied, exclusions, residual risks and a change-control method if the revision advances before acceptance.

Under-rider-specific hazards

• foot, hand or cleaning-tool trapping while the vehicle enters beneath a cart;
• cart tilt or structural damage during lifting;
• load overhang contacting a person or machine first;
• departure with incomplete coupling or uncoupling in motion;
• cart rollaway or unexpected restart during manual recovery;
• cart or load obstruction of a protective sensor field;
• charging and parking obstructing egress, firefighting access or forklift traffic.

Controls can include speed zones, restricted access, physical guards, cart stops, warnings, lockout, inspection, training and controlled software/layout changes. Qualified parties should validate safety functions against actual Thailand legal and plant requirements. A component declaration is not system acceptance.

Fleet sizing must include variation, exceptions and charging

Multiplying average travel time by move count underestimates capacity. The cycle contains dispatch waiting, cart search, alignment correction, engagement, lift, doors, intersections, handoff confirmation, empty-cart repositioning, charging and exception recovery. Evaluate the upper side of the time distribution and production peaks, not only a median.

Stress at least normal production, peak windows, changeovers, one robot unavailable, one charger unavailable, route closure, congested forklift traffic, wrong cart placement and canceled moves. High planned utilization can create a nonlinear queue when one vehicle stops.

Charging design covers charger count, simultaneous demand, queueing, contact cleaning, restart after a power interruption, fire precautions and battery maintenance. A catalogue runtime or charging time is one input. Only a duty-cycle test with the site’s routes and loads can support an operational plan.

Define system states before integrating MES, WMS or ERP

An API that sends move commands is insufficient. Define accepted, assigned, cart reserved, pickup attempted, coupled, in transit, delivered, canceled, held, manually handled, retried and closed. For each transition, state which system is authoritative and how duplicates, late responses and retries are reconciled.

VDA announced VDA 5050 version 3.0 on 20 April 2026. Its public release discusses master control for mixed fleets, zones and shared paths, local-language error information and energy-saving actions. Standard communication can reduce integration ambiguity, but it does not by itself guarantee plug-and-play interoperability, safety, business semantics or mechanical compatibility. Confirm vendor version, optional fields, extensions, map conventions, conformance testing and upgrade governance.

Error messages should tell the operator the location, cart, condition, safe first action and escalation point in the operating language. Role-based access must prevent a helpful message from becoming permission to bypass a safety control. Remote support, logs, clock synchronization, personal-data treatment and cybersecurity belong in acceptance testing.

What to include in an under-rider AGV RFP

For each requirement, specify mandatory/desirable status, test method, evidence, owner and acceptance location.

Mechanical and cart interface

• 3D envelope and entry/exit trajectory;
• total load, center-of-gravity range, offset load, lift-point reaction and reinforcement;
• cart tolerance, caster condition, wear, deformation and foreign-object limits;
• coupling, lift, hold, release, cart presence and wrong-cart detection;
• manual release, disabled-vehicle recovery, tools and staffing.

Route, safety and equipment interface

• applied standard edition, risk-assessment scope and safety-function list;
• loaded/unloaded speed zones, stopping, crossings, occlusion and side risk;
• fail-safe behavior for doors, shutters, PLCs, beacons and emergency stops;
• behavior on communication loss, localization loss, sensor contamination and low battery;
• evacuation, fire, power loss, maintenance, update and layout-change procedures.

Performance, data and lifecycle

• measured distribution from request to delivery and its conditions;
• definitions for completion, retry, intervention and stop reason;
• fleet, charging, spare vehicle, spare parts and service assumptions;
• API, state model, timestamps, logs, backup and restore;
• Thailand support, training languages, consumables, service contract and end-of-life handling.

FAT should create failures, not just display a finished machine

FAT is the opportunity to expose mechanical, control, safety and recovery defects before site installation. Use production-representative carts, casters, loads, IDs and host signals. If substitutes are used, record the difference and repeat the affected test during SAT.

Representative FAT cases include minimum- and maximum-clearance carts; empty, normal, maximum and allowed offset loads; XY and angular parking errors; missing and wrong cart IDs; incomplete lift or coupling; front and side intrusion; blocked sensors; network delay and loss; duplicate and canceled orders; low battery; charger contention; emergency stop; manual release; recovery; local-language alarms; permissions; logs; clock sync; software versions; configuration backup and documentation handover.

Every case needs preconditions, input, procedure, expected result, measurement, log reference, pass/fail, unresolved item and retest date. “It ran ten times” is weak evidence unless the tested cart, load, floor and configuration are traceable.

SAT proves the system on the actual Thailand factory floor

SAT addresses conditions FAT could not reproduce: worst floor points, fire doors, live intersections, forklifts, people, Wi-Fi, equipment signals, cart variation, cleaning condition, temperature and shift change. Safety protections must not be relaxed to protect production schedules; define approval authority for any change.

Acceptance should distinguish a successful delivery, a correct safe rejection, recovery by automatic retry and manual intervention. Counting all safety stops as simple failures pressures people to suppress stops; excluding every stop as “safe operation” hides poor operability. Keep cause codes and recovery time together. Site-approved production demand and peak patterns should set performance gates; this article supplies no universal pass percentage.

Hypothetical ROI: retain the case where labor alone does not pay

The following is one fictional worldline, not a market average, customer result, quotation or guarantee.

• One internal route, 18 cart moves per shift, two shifts per day, 250 days per year.
• Current handling time is seven minutes per move.
• Planning labor cost is THB 180 per hour.
• Seventy percent of walking/handling time is avoidable after stabilization.
• Annual robot-system cost is assumed at THB 420,000, including annualized equipment/integration, service, charging electricity and routine maintenance.

Gross annual time is 18 × 2 × 250 × 7 ÷ 60 = 1,050 hours. Potential released labor value is 1,050 × 70% × THB 180 = THB 132,300/year. The simple difference is THB 132,300 − THB 420,000 = −THB 287,700/year.

Direct labor alone does not justify this worldline. Proceed only if separately evidenced benefits—line-starvation avoidance, ergonomics, traceability, space, schedule stability or productive redeployment—close the gap, or redesign the scope and cost. Do not monetize time unless headcount, overtime, outsourcing or productive redeployment can actually change, and do not double-count it under another benefit.

A clearly hypothetical 90-day under-rider AGV PoC

The following 90-day PoC is a planning example. Scope and gates require site approval; they are not universal acceptance values.

Days 0–30: freeze fit conditions and baseline

Limit scope to one route, one cart family and controlled loads. Identify every cart; measure clearance, center of gravity, deflection and casters. Survey floor, crossings, doors, communications, charging, evacuation and recovery. Baseline demand, time, queues, intervention and shortage effect. Agree on risk assessment, training, permissions, stop rules and change approval.

If cart strength or tolerance cannot be explained, decide on modification, fixture, another handling concept or narrower scope before increasing cycles.

Days 31–60: run loaded cycles and forced exceptions

Repeat empty, normal, maximum and allowed offset-load cases. Trigger cart misplacement, wrong cart, incomplete coupling, blocked route, door delay, network loss, localization loss and low battery. Perform emergency stop, manual release, recovery and restart by role. Reconcile move state, alarms, operator action, recovery time and host-system records.

Classify failures as design, cart, floor/equipment, instruction data, operating deviation or training. Do not leave unexplained stops in a permanent “other” bucket.

Days 61–90: production-like operation and decision

Run normal and peak windows, repeat edge cases, verify charging queues and contingency with one robot or route unavailable. Transfer inspections, cleaning, backup, updates and cart checks to site owners. Replace ROI assumptions with observed data and exclude unrealized benefits. Decide among scale, modify, narrow or stop.

A useful scorecard links cart conformity, delivery-time distribution, safe rejection, recovery, host reconciliation, maintenance ownership and updated economics to evidence. An unresolved safety risk stops scale-up regardless of throughput.

Thailand investment context is not project eligibility

A Thailand BOI first-half 2026 release identifies THB 17.2 billion for machinery upgrades, digital technology adoption and integration of automation and robotics. A separate Q1 2026 BOI release states Machinery, Automation and Robotics approvals of THB 8,081 million across 38 projects, while a related stated application measure is THB 7,071 million across 61 applications. Approvals, applications, groupings and periods differ; they should not be added or presented as one growth calculation.

The releases provide context for automation investment but do not establish that a particular AGV project qualifies for incentives. Confirm eligibility, timing and cost treatment directly with BOI or qualified advisers, and preserve an investment case that does not depend on an unconfirmed incentive.

Common implementation failures

Selecting by payload alone: use offset loads, cart deflection, lift reactions and dynamic stops in FAT/SAT.

Measuring one “representative” cart: establish cart IDs, inspections and rejection limits.

Sizing from an average cycle: include peaks, doors, crossings, charging and recovery.

Delegating all safety to the vehicle vendor: assign plant, EHS, production, maintenance, IT and integrator responsibilities.

Designing the PoC as a successful demo: intentionally create failures, classify them, recover, log and retest.

FAQ: under-rider AGV and cart automation selection

What is an under-rider AGV?

It is a low-profile driverless vehicle that enters beneath a cart or shelf and couples through a lift, pin, hook or engineered interface. It can reduce transfer handling, but the cart envelope, strength, load center, coupling tolerance, floor, safety and recovery must all be validated.

Can a low-profile AGV use existing carts without modification?

Sometimes, but not by height alone. Check braces, caster sweep, brakes, bottom-plate deflection, lift points and unit-to-unit variation. A supplier statement about a model cannot replace measurement and loaded testing of the actual carts.

What should be decided first for roll-cage transport automation?

Define the cart family, load envelope, route, move demand, empty/full and quality states, parking position and reject criteria. Then survey carts and define the boundary between normal and unsafe or incompatible conditions.

How are FAT and SAT different for an under-cart vehicle?

FAT exposes mechanical, control and exception defects before shipment in a supplier/integration environment. SAT confirms performance on the actual floor with people, traffic, communications, doors, equipment, cart variation and production peaks. Substitute FAT conditions must be repeated at SAT.

Is a claim of ISO 3691-4 compliance enough for AGV selection?

No. Ask for the applied edition, system boundary, operating-zone assumptions, exclusions, residual risks and test evidence. Because the third edition is still a DIS, include revision monitoring and contractual change control.

Does VDA 5050 make mixed fleets immediately plug-and-play?

It can support standardized communication, but implementation versions, optional fields, extensions, maps, business states, safety and mechanical interfaces still require testing. Define conformance and incident ownership.

Can under-rider AGV ROI be calculated from labor savings only?

It can be modeled, but labor alone may not pay. Use realizable headcount, overtime, outsourcing or productive redeployment, and evidence other benefits separately. The fictional case in this article is negative on direct labor alone.

What should pass a 90-day PoC?

Evaluate cart conformity, safe rejection, exception recovery, peak and charging behavior, host integration, maintenance transfer and updated economics—not only successful move count. Unresolved safety risk is a stop condition.

Conclusion: select the cart-and-vehicle system, not a low body alone

Vehicle height and payload are only the entry screen for an under-rider AGV. A defensible decision measures each cart’s envelope, load center, reinforcement, casters and coupling tolerance; treats floor, doors, traffic, communications, charging and recovery as part of the system; and defines evidence before purchase. FAT should create controlled failures, SAT should expose real-site variation, and the 90-day PoC should retain an honest option to modify, narrow or stop. That is how a promising demo becomes supportable production transport.

TOMAS TECH can support Thailand factories with cart and route surveys, under-rider AGV requirements, vendor comparisons, FAT/SAT scenarios, host-system integration and 90-day PoC design. You can contact us while you are still checking whether existing carts are suitable; a product decision is not required first.

References

• ISO, ISO 3691-4:2023: https://www.iso.org/standard/83545.html
• ISO, ISO/DIS 3691-4 draft third edition: https://www.iso.org/obp/ui?_escaped_fragment_=iso%3Astd%3Aiso%3A3691%3A-4%3Adis%3Aed-3%3Av1%3Aen
• VDA, VDA 5050 version 3.0 release: https://www.vda.de/en/press/press-releases/2026/260421_PM_VDA_5050_EN
• OMRON, OL-450S: https://robotics.omron.com/products/mobile-robots/ol-series/ol-450s/
• Sharp, TYPE S: https://www.sharp.co.jp/business/agv/products/slim/
• Meidensha, ultra-low-floor lift AGV: https://www.meidensha.co.jp/products/logistics/prod_01/prod_01_10/
• Mobile Industrial Robots, MiR600 Shelf Lift: https://mobile-industrial-robots.com/products/applications/mir-shelf-lift-600
• Thailand BOI, first-half 2026 release: https://osos.boi.go.th/EN/news/2430/Thailand-Secures-43-6bn-1H-2026-Investment-Surge-as-Big-Tec/
• Thailand BOI, Q1 2026 release: https://www.boi.go.th/index.php?_module=news&from_page=press_releases&language=en&page=press_releases_detail&topic_id=138788
• Meidensha AGV site, low-platform lineup: https://agv.meidensha.co.jp/agv/low-platform.html

This article provides general procurement and system-design information, not safety, legal, tax or investment-incentive advice. Qualified parties must validate applicable standards, laws, risk controls, cart strength and acceptance criteria against the actual site. Confirm product data in the supplier’s current contract documents. The ROI and 90-day PoC figures are assumptions, not market averages, customer outcomes, quotations or guarantees.