AGVs are the default choice when a plant runs the same load, on the same path, dozens of times per shift, with a ceiling capacity that reaches about 5.5 tonnes per vehicle, well above the typical AMR envelope [S4].
AMRs, by contrast, are sized for lighter, more varied work, with payload ratings commonly landing between 0.11 and 2.2 tonnes, and they justify that lower capacity with autonomous navigation that does not require floor infrastructure [S4][S1]. The decision is not which robot is "better" but which constraint binds first: load and route rigidity, or layout volatility and human traffic.
What "fixed-route heavy" actually demands from the vehicle
AGVs are unmanned transport systems that follow predefined routes guided by magnetic tape, embedded wires, QR codes, or optical lines, which is exactly the behaviour a heavy-load, high-cycle duty cycle rewards [S5]. Because the path is encoded in the floor rather than the vehicle, AGVs can be built around larger chassis, heavier drive trains, and bigger batteries without paying a navigation-software tax; AGVs are widely used in automotive plants moving engines and car bodies, in food and beverage for raw-material and finished-goods transfer, and in pharmaceutical sterile material transport [S3].
That same rigidity, however, is the AGV's ceiling: a fixed route cannot be rerouted around a dropped pallet or a stalled forklift, and the unit will stop and wait for the obstruction to clear rather than find an alternate path [S2][S6]. Reconfiguration also carries a real cost, because changing the route usually means re-laying magnetic strips, moving reflectors, or re-cutting slots for inductive wires, which is why AGV fleets are sized to the original layout and stay there [S5].
Why AMRs trade payload for navigation freedom
AMRs navigate dynamically using SLAM, LiDAR, 3D cameras, UWB, and IMU sensor fusion, and they plan paths in real time rather than reading them off the floor [S3][S5]. On a route change, no civil work is needed: the map is updated in software, the fleet manager re-assigns missions, and the fleet keeps moving, which is why deployment is measured in weeks for AMRs and in months for AGV retrofits [S2].
That flexibility comes with two penalties. The sensor and compute stack adds cost and mass that competes with payload, so AMR payloads cluster in the 0.11–2.2 t band documented in cross-vendor surveys [S4]. And AMR safety relies on integrated scanners that detect people and react, rather than on physical separation, which works in a mixed-traffic warehouse but is not a free pass on a 30-tonne line-side tug route where the right answer is still a heavy AGV on a dedicated lane [S5][S1].
Decision matrix: AGV versus AMR on the binding criteria
Stacked against the four criteria that actually drive a spec, the picture is sharp. On load capacity, AGVs reach about 5.5 t and AMRs about 0.11–2.2 t, roughly a 5x ceiling advantage for AGVs [S4]. On route stability, AGVs win where the path does not change, and lose badly when it does, because every re-route is a floor-work event [S5][S2]. On deployment speed, AMRs map and go in days with no infrastructure; AGVs need magnetic tape, wires, or QR codes installed and validated before the first load moves [S2][S3]. On obstacle handling, an AGV detects and stops, an AMR detects and reroutes, which is the single biggest behavioural gap between the two classes [S6][S2].
A useful rule of thumb: if the route, load, and cycle are stable for at least 3–5 years, the AGV's lower unit cost and higher capacity win; if any of those three move more than quarterly, the AMR's zero-infrastructure reconfiguration wins even at a lower payload [S3][S5]. For mixed-fleet sites, see the related spec work on AGV onboard controller and traction motor architecture, options, and sizing, which covers the drivetrain side of the heavy-load decision.
Use cases where the AGV is the right tool
Three patterns keep showing up in vendor and integrator write-ups. First, line-side replenishment in automotive final assembly, where engines, dashboards, and body-in-white subassemblies run the same 200–800 m loop for 15+ years and AGVs handle the tonnage that AMRs physically cannot [S3][S1]. Second, pallet and roll handling in paper, steel, and beverage mills, where the route is laid into the building and the loads run 2–5 t per move, again well inside the AGV envelope and outside the AMR envelope [S3][S4]. Third, cleanroom and sterile suites in pharmaceutical plants, where magnetic-strip or inductive-guide AGVs limit particulate generation versus a free-navigating AMR's wheel dust and sensor wash-down risk [S3].
Each of these is the inverse of the AMR sweet spot: high-mix e-commerce picking floors, hospital meal and medication delivery, and retail back-of-store replenishment, where the route is "any aisle, any time" and a 250 kg payload is plenty [S3][S2].
Limits, failure modes, and integration gotchas
The AGV's biggest operational risk is a single point of physical guidance: a peeled magnetic strip, a buried inductive wire break, or a reflector knocked out of alignment will stop the whole lane until it is repaired, and AGV downtime is more often infrastructure-driven than vehicle-driven [S5]. AMRs trade that failure mode for two of their own, namely dependency on environmental sensing (a featureless corridor, mirror walls, or heavy dust degrades SLAM) and a higher per-unit cost that is justified only when the route mix actually changes [S3][S2].
Integration is also asymmetric. AGVs are usually specified and installed by the vehicle vendor or a systems integrator, with WMS/MES hooks bolted on after; AMRs typically integrate natively with WMS/MES/ERP because the fleet manager is software-defined and the vehicles are downstream of it [S3][S5]. For a deeper look at the wider mobile-robot category and where the dividing lines sit, the AGV robot primer covers the same architecture at the component level.
Sourcing checklist for a fixed-route heavy-load AGV project
Lock four numbers before vendor selection. First, peak payload per move, including a 1.25–1.5x design margin, and check that it sits inside the AGV's published rating rather than at the very top of it [S4]. Second, route length, number of nodes, and intersection count, because those drive the battery sizing, the fleet count, and the traffic-management logic more than the vehicle spec sheet does [S5]. Third, the floor and environment: magnetic tape versus inductive wire versus QR codes has very different install cost, cleanroom compatibility, and maintenance burden, and the choice is not vendor-neutral [S5][S2]. Fourth, safety category, which in most jurisdictions means dual-channel safety scanners on the vehicle plus physical separation or light curtains at intersections, with the standard regime typically set by ISO 3691-4 for driverless industrial trucks.
On the financial side, AGVs demand a higher initial outlay and a longer ROI than AMRs, and the literature is consistent that this is acceptable only when the underlying process is stable and repetitive enough to drive high utilisation [S2][S5].
For adjacent engineering decisions outside mobile robotics, see the comparison-driven spec work on AC induction versus DC motors on a hydraulic power unit and on 230 VAC versus 400 VAC servo drives for small axes, both of which follow the same criteria-first, opinionated-pick pattern. Track these signals next: any vendor releasing a >5 t AMR with safety-rated SLAM (would re-open the heavy-load AMR case), and any Tier-1 automotive OEM publishing a brownfield AGV-to-AMR migration cost benchmark (would set a real number on the reconfiguration penalty).
For the relevant spec sheets and selection criteria, see fixed gas detector, and pressure transmitter.