Autonomous Mobile Robot procurement in mid-2026 is decided by five hard gates — payload class, navigation architecture, fleet-manager openness, safety rating, and floor-condition tolerance — and skipping any one of them routinely turns a quoted 8-week deployment into a 6-month integration project [S2].
The category sits between two older families in the automation taxonomy: the line-following AGV robot world, which still dominates heavy-pallet moves above 1,500 kg, and the vision-led AMR robot tier, which owns the 50–1,500 kg mid-range where 80% of new greenfield cells now land. Picking the wrong tier is the single most expensive line item in an AMR spec sheet [S1].
Payload × Reach × Floor Class: The First Gate
Payload is non-negotiable: spec the gross vehicle weight (chassis plus battery plus rated load) at 1.25× the heaviest tote you actually run, not the catalogue peak, because end-of-shift battery sag shifts the centre of gravity and derates real lifting capacity by 8–12% on lithium-iron packs [S2].
Reach divides into three practical bands — under 1,000 mm (bin/cart tug), 1,000–1,800 mm (roller-top transfer to conveyors), and above 1,800 mm (pallet lift, where the unit straddles a EUR/CHEPP pallet on forks). Below 1,000 mm of lift height, differential-drive chassis dominate; above 1,800 mm, the spec must also call out mast type (single-stage telescopic vs triplex) and residual capacity at full elevation, which on most 1.6 t pallet AMRs drops to 800–1,000 kg [S1].
Floor class is the third axis most RFQs skip: expansion joints wider than 8 mm, grated mezzanine decks with 25–30 mm openings, and resin-coated slopes above 3° will each knock 15–30% off vendor-quoted speed, and on some collaborative robot-class chassis they trigger a hard shutdown. Insist on a measured floor survey — joint width, slope map, coefficient of friction (target μ ≥ 0.45 dry) — before sign-off [S2].
Navigation Stack: SLAM vs Line-Follow vs Hybrid
Three navigation architectures compete in 2026: 2D laser SLAM with reflector augmentation, natural-feature 3D SLAM (vision + LiDAR fusion), and the older magnetic-tape or QR-floor line-follow stack still common in AGV robot retrofits [S1].
2D laser SLAM with reflectors remains the workhorse for warehouses above 5,000 m² because localisation drift stays inside ±10 mm once reflectors are placed at 8–12 m spacing; it is also the easiest stack to re-route when a floor layout changes, since reflectors are passive and the map is editable. Natural-feature 3D SLAM removes the reflector install cost (typical saving 8–15 € per reflector, hundreds per site) but adds 2–4 weeks of mapping labour and demands a stable visual scene — empty white-wall aisles defeat it [S2].
Hybrid stacks, which fall back from natural-feature SLAM to contour-follow when lighting or scene clutter degrades, are the fastest-growing segment in 2026 OEM releases; they trade a 10–20% premium in sensor cost for a measurable drop in mean-time-between-intervention. The Intel Autonomous Mobile Robot reference platform packages SLAM, classification and planning as pre-validated ROS 2 modules, and is the closest thing to a vendor-neutral benchmark a spec writer can cite [S2].
Fleet Manager Openness and WMS/MES Hooks
Fleet-manager openness is where most 2025–2026 RFQ disappointments are born: a closed fleet manager that exposes only REST endpoints to a warehouse management system locks the site to one vendor's spare-parts, one vendor's firmware cadence, and one vendor's pricing on every future robot add [S1].
Three integration tiers are now common in vendor literature. Tier 1 is a vendor-proprietary fleet manager with documented REST/OPC UA endpoints — adequate for a single-vendor site of under 30 units. Tier 2 is a vendor-neutral fleet manager (VDA 5050-compatible) that lets a mobile crane of mixed-brand AMRs share traffic; this is the 2026 default for any greenfield above 30 units. Tier 3 is a fully open-source stack (e.g. ROS 2 + Open-RMF), with the lowest licence cost and the highest integration engineering hours — typical 6–10 person-months of bespoke work for a 50-unit site [S2].
The other half of the WMS/MES question is the call-up protocol. VDA 5050 is the de-facto European standard for fleet-to-vehicle orders and is cited in 2026 OEM datasheets alongside the older ISA-95 pick-list conventions. For brownfield sites running SAP EWM or Manhattan Active, demand a vendor-supplied adapter or a documented JSON schema for the order, state and visualization messages — without it, the WMS team owns the integration risk [S2].
Safety Rating, Battery Chemistry and TCO
Safety rating for mobile robots in 2026 maps to ISO 3691-4:2020, which splits the fleet into classes by rated payload, maximum speed, and whether the unit operates in a shared space with pedestrians. The Performance Level required on the safety-rated laser scanner and bumpers is PL=d per EN ISO 13849-1, with a Safety Integrity Level that tracks the same boundary. Specifying ISO 3691-4 compliance in the RFQ is no longer optional in EU plants — most insurers will not cover a scara robot cell that has a wandering AMR in it without it [S1].
Battery chemistry decides the duty cycle. Lithium iron phosphate (LFP) is the 2026 default for new fleets because it tolerates opportunity charging at 1C, supports 4,000–6,000 full cycles to 80% state-of-health, and removes the cobalt-nickel supply exposure that has dragged lead times on NMC packs in past quarters. For cold-chain sites (below 0 °C), specify self-heating LFP packs; standard NMC cells lose roughly 30% of usable capacity at –10 °C [S2].
The "service and spares" line is the one buyers underestimate: budget 6–9% of fleet capex per year for drive-wheel replacement, scanner window cleaning, and battery refresh around year 6 [S1].
Decision Matrix: AMR vs AGV vs Fixed Automation
Across the five gates, the practical decision matrix for 2026 reads as follows. AMRs win in the 50–1,500 kg range, on changing routes, in mixed-traffic aisles, and in any site where adding a 10th unit must not require a control-cabinet retrofit [S1].
Fixed automation — conveyors, articulated robot cells, stacker crane AS/RS — still wins on throughput above roughly 200 moves per hour on a single lane, where the AMR's 60–80 moves/h per unit cap is binding. The hybrid pattern now common in 2026 greenfields is AMR tug-train delivery to fixed pick-stations, with pallet rack vs AS/RS zones reserved for high-velocity SKUs.
Three rules of thumb to carry into vendor demos: 1) refuse any demo that runs in a clean, empty hall — demand it on your floor with your totes; 2) measure mean-time-between-intervention in hours, not in moves; 3) require a reference site that is at least 12 months old, not the 6-week-old launch customer the vendor wants to parade.