Autonomous Mobile Robot

What is an Autonomous Mobile Robot

An autonomous mobile robot is a wheeled or tracked industrial vehicle that moves materials through a working environment under its own decision-making, without a driver and without fixed guide infrastructure such as magnetic tape or buried wire. The defining capability is autonomy: the robot perceives its surroundings with onboard sensors, maintains an internal map, decides its own route, and adapts that route in real time as people, forklifts, and other robots cross its path. This separates an AMR from a conveyor (a fixed mechanical path) and from a classical AGV (a fixed virtual path the vehicle is not allowed to leave).

Functionally, an AMR is built from four subsystems. First, a drive base: a wheeled chassis (differential-drive, tricycle, or omnidirectional mecanum configurations are common) with motors, motor controllers, encoders, and suspension. Second, a perception stack: one or more 2D safety LiDAR scanners for the protective field, plus 3D cameras or 3D LiDAR for volumetric obstacle detection and pallet recognition. Third, a compute and navigation core: an industrial computer running the SLAM, localization, path-planning, and obstacle-avoidance algorithms. Fourth, a payload interface: a flat top deck, a conveyor or roller top, a lift mechanism for under-shelf carrying, or a mounted robot arm for mobile manipulation.

The lineage of the AMR runs through the automated guided vehicle. The first AGV appeared in 1953 as a tow tractor following a wire in the floor, and for decades industrial mobile automation meant laying physical guide paths. The breakthrough that produced the modern AMR was practical, low-cost SLAM: by combining affordable 2D LiDAR, more powerful embedded computing, and probabilistic localization algorithms, a vehicle could finally build and use its own map rather than depend on infrastructure. Commercial free-navigating mobile robots reached the warehouse market in the 2010s, and the category expanded rapidly with the surge in e-commerce fulfilment and the long-running shortage of warehouse labour.

The scale of the market reflects that demand. Independent market studies value the global AMR segment in the low single-digit billions of US dollars as of 2024, with warehousing and logistics the single largest application, and forecast double-digit annual growth through the early 2030s as fulfilment centres, manufacturing plants, and hospitals continue to automate internal transport. Because the dominant use case is moving the same load between the same points many times a day, AMRs deliver their return on investment through sustained labour displacement and higher asset uptime rather than through any single dramatic capability.

Four engineering metrics govern whether an AMR fits an application: payload capacity, travel speed, positioning and docking accuracy, and the safety performance level of its obstacle detection. Around these sit the practical realities of battery and charging strategy, navigation method, and fleet-management software. The rest of this guide decodes each of these, because the most expensive AMR mistakes are made not on the showroom floor but in a specification document that confused free-driving speed with throughput, or that ignored fleet interoperability and locked a buyer into one vendor for a decade.

AMR Types and AMR vs AGV

Mobile robots divide first by how they navigate, and second by what they carry. The most consequential split, and the one that causes the most confusion in tenders, is between the fixed-path automated guided vehicle and the free-navigating autonomous mobile robot. The two share a chassis lineage but behave very differently around obstacles and around layout change. The table below contrasts them across the decisions an engineer actually has to make.

The distinction is no longer absolute. A vehicle can carry SLAM sensors yet be configured to follow a fixed virtual route, and many vendors now sell the same hardware as either an AGV or an AMR depending on the control mode. For this reason the North American safety standard ANSI/RIA R15.08 deliberately uses the umbrella term industrial mobile robot (IMR) to cover both. The practical test remains behavioural: if the vehicle stops and waits when it meets an obstacle, it is acting as an AGV; if it reroutes around the obstacle on its own, it is acting as an AMR.

By load-handling format, AMRs fall into several families. Tote and case carriers have a flat or shelved top deck and move bins, cartons, and totes in fulfilment and light manufacturing. Pallet movers carry full euro or block pallets, often with a fork or lift interface, in the 500 to 1,500 kg class. Under-shelf (goods-to-person) robots drive beneath a movable rack, jack it up, and carry the entire shelf to a picking station, the model popularized by large e-commerce fulfilment centres. Tugger or tow AMRs pull a train of carts along a milk-run. Conveyor-top and lift-top AMRs automate the handoff to fixed conveyors and machines. Mobile manipulators add a collaborative robot arm on the base for picking or machine tending.

By drive geometry, the common configurations are differential drive (two independently driven wheels plus casters, simple and able to turn on the spot), tricycle and Ackermann steering (for higher-speed or heavier vehicles), and omnidirectional mecanum or steered-wheel layouts that can translate sideways and rotate independently. Omnidirectional bases shine in tight, congested spaces and in precise lateral docking, at the cost of mechanical complexity and wheel wear. Differential drive remains the workhorse for general transport because it is robust, cheap, and well understood. Drive choice constrains the minimum aisle width and the achievable docking strategy, so it should be decided alongside payload, not after it.

Navigation Technologies

Navigation is the heart of mobile robotics, and it is where AMRs earn their premium over AGVs. The job has three parts: localization (where am I), mapping (what does the world look like), and path planning with obstacle avoidance (how do I get there safely). The dominant modern method solves all three with SLAM. Legacy guided methods solve only localization, and only along a pre-laid path. The table compares the main navigation approaches across accuracy, infrastructure cost, and flexibility.

Natural-feature SLAM is the default for AMRs. A 2D safety LiDAR sweeps a horizontal plane many times per second, returning the distance to walls, racks, machines, and fixed structure. The robot matches that live scan against a stored occupancy map to estimate its pose, while a planner computes a collision-free route and an avoidance layer handles dynamic objects. Because the map is built from features that already exist in the building, no tape, wire, or reflector has to be installed, and the same map can be edited in software when the layout changes. The trade-off is that free-driving path accuracy is looser, on the order of plus-or-minus 10 to 50 mm, which is ample for transport but not for precise handoff without a local docking aid.

Docking accuracy is therefore handled separately from cruising. When an AMR approaches a conveyor, charger, or pick station, it switches to a local reference: a recognized shelf leg, a reflector pair, or a printed marker, and closes the loop to a much tighter tolerance. Production datasheets quantify this directly; the MiR250, for example, specifies plus-or-minus 3 mm on both X and Y at a docking marker. Where even that is insufficient, the mobile base only positions coarsely and a separate mechanism, a robot arm or a passive cone-and-funnel guide, performs the final sub-millimetre alignment.

Guided methods persist where the route truly never changes. Inductive wire buried in the floor gives roughly plus-or-minus 1 mm repeatability and is almost indestructible, suited to permanent heavy-haul routes, but installation means cutting the floor and any change is costly. Magnetic tape is the cheapest entry point at about plus-or-minus 5 mm, but tape is exposed to forklift traffic and wear. Laser triangulation off wall-mounted reflectors reaches plus-or-minus 1 to 5 mm and needs clear sightlines to the reflectors. QR or visual-code grids on the floor are the navigation basis of many under-shelf goods-to-person fleets, where dense, regular codes let thousands of robots localize cheaply in a structured grid.

Most warehouses end up with a hybrid: SLAM for the open transport space and a local docking aid at each critical handoff point. The selection question is rarely SLAM versus tape in the abstract, but rather how much the layout will change over the asset's life, how many handoff points need tight accuracy, and whether the facility can tolerate any floor-mounted infrastructure at all. The more a layout is expected to evolve, the more decisively SLAM wins on lifetime cost.

Drive, Battery, and Safety Systems

Behind navigation sit three subsystems that determine whether an AMR can actually run a shift safely: the drive and motion hardware, the energy and charging strategy, and the functional-safety chain. These are where datasheets either earn trust or quietly hide a limitation, so they deserve as much scrutiny as the headline payload figure.

Drive and motion. Most AMRs use brushless DC or servo wheel motors with integrated encoders, driven by closed-loop controllers that the navigation core commands. Differential-drive bases zero-turn and are mechanically simple; omnidirectional bases add lateral motion for tight docking. Wheel and floor condition matter more than buyers expect: polyurethane drive wheels need a clean, level, dry floor, and a 1 to 2 percent gradient or a wet patch can change traction and braking distance, which feeds directly into the safety field calculation. Suspension keeps all drive wheels loaded over expansion joints and minor floor unevenness so localization and traction stay consistent.

Battery and charging. Modern AMRs run on lithium-ion packs, generally LFP (lithium iron phosphate) for thermal robustness and cycle life or NMC for energy density. Lithium chemistry tolerates frequent partial top-ups without the memory effect or sulphation that limited older lead-acid fleets, which makes opportunity charging practical: the robot snatches short charges during natural idle gaps rather than being parked for hours. The metric to read is the charge ratio. The MiR250, for instance, quotes up to 1:16, meaning roughly 10 minutes of charging buys about 2 h 40 min of runtime, and rates a minimum of 3,000 full cycles. With opportunity charging and enough chargers, a fleet can approach round-the-clock availability without spare-battery swapping.

Functional safety. The safety-critical function of an AMR is detecting people and obstacles in time to stop or slow. This is the job of the 2D safety laser scanner, a Type 3 device under IEC 61496 such as the SICK microScan3 or nanoScan3, configured with protective and warning fields that shrink and grow with speed. Such a scanner achieves Performance Level d (PLd) per ISO 13849-1, equivalent to SIL 2 per IEC 62061, which is the typical rating for AMR obstacle detection. The robot reduces speed when an object enters the warning field and triggers a safe stop when the protective field is breached. Emergency-stop buttons, safe-torque-off in the drives, and audible and visual warning signals complete the chain. The whole machine is then assessed against ISO 3691-4 for CE marking in Europe and ANSI/RIA R15.08 in North America, both anchored on ISO 12100 risk assessment.

The table summarizes the safety and power building blocks an engineer should verify on any AMR datasheet, with the standards each maps to.

Key Specification Parameters

An AMR datasheet can list thirty lines, but seven parameters drive nearly every selection decision: payload, maximum speed, positioning and docking accuracy, footprint, runtime and charge ratio, safety performance level, and fleet-interface support. Each is explained below, with verified reference figures from published manufacturer datasheets so the ranges are concrete rather than invented.

Payload is the maximum mass the robot is rated to carry, and it sets the whole class of vehicle. Published examples bracket the field: the OTTO 100 is rated for up to 150 kg, the MiR250 for 250 kg, the OTTO 600 and MiR600 for 600 kg, the MiR1350 for 1,350 kg, and the OTTO 1500 family for heavy pallet loads in the 1,500 kg class (an updated variant rated up to about 1,900 kg). Always rate against the heaviest realistic load with margin, and confirm the load's centre of gravity stays inside the manufacturer's stated envelope, because a tall or off-centre load forces the robot to lower its safe speed.

Maximum speed on a datasheet is the free-driving figure on an open path. Common values run from about 1.2 to 2.0 m/s; the MiR250 and OTTO 100 cite up to 2.0 m/s, while heavier units such as the MiR1350 cite about 1.2 m/s. Crucially, top speed is not throughput. The robot automatically slows near people, in narrow aisles, and on turns, and a large share of cycle time is spent docking, waiting in traffic, and charging. Two AMRs with identical top speeds can differ greatly in real pallets-per-hour depending on layout and fleet coordination.

Positioning and docking accuracy separates cruising from handoff, as Chapter 3 covered. Free SLAM driving holds roughly plus-or-minus 10 to 50 mm; docking at a marker tightens to about plus-or-minus 3 mm (MiR250) using a local reference. Specify the accuracy you need at the handoff point, not the open floor, and budget a docking aid if your interface is tight.

Footprint and height determine which aisles the robot fits and how it passes under shelving. The MiR250, for example, has an 800 by 580 mm footprint at 300 mm height, low enough to drive under and lift movable racks. A small footprint with a tight turning radius is decisive in retrofit warehouses where aisles were never designed for automation.

Runtime and charge ratio together describe availability. Published runtimes reach roughly 13 hours with full payload and over 17 hours unladen on a single charge for a 250 kg-class robot, but in practice opportunity charging matters more than a single deep cycle. Read the charge ratio (up to 1:16 on the MiR250) and the rated cycle life (3,000 cycles or more) to model how the fleet behaves across a continuous operation, not just one shift.

Safety performance level should appear explicitly. Expect the obstacle-detection function at PLd per ISO 13849-1 (SIL 2 per IEC 62061), and the machine certified to ISO 3691-4 (CE) and ANSI/RIA R15.08, often with ANSI/ITSDF B56.5 and EN 60204-1 also cited. Fleet-interface support is the seventh parameter and the one most often omitted: confirm whether the robot exposes a VDA 5050 compliant interface so it can be coordinated with mixed-brand fleets through a universal master control, and confirm the integration path to your WMS or MES.

Selection Decision Factors

To turn the preceding five chapters into a model and a tender, follow the decision sequence below. Most AMR selection errors come not from one wrong number but from settling a later decision before an earlier one, for example fixing on a vehicle before mapping the real traffic flow. These eight steps can serve as a fixed RFQ template.

1. Define the transport task first: map the actual flows (from-point, to-point, frequency, load type, peak rate) before looking at any robot. Throughput targets and aisle widths come from the flow, not the datasheet.
2. Payload and load format: pick the class from the heaviest realistic load plus margin (150 / 250 / 600 / 1,350 / 1,500 kg are common rungs), and choose the top interface: flat deck, conveyor, lift for under-shelf, fork for pallets, or a mobile manipulator.
3. Navigation method: default to natural-feature SLAM for any facility whose layout will evolve; consider a fixed-path guided vehicle only for a permanent, high-volume single loop where the lowest cost per vehicle dominates.
4. Accuracy at handoff: separate cruising tolerance (plus-or-minus 10 to 50 mm is fine) from docking tolerance (specify plus-or-minus 3 mm or tighter where a conveyor or arm receives the load), and budget the docking aid that delivers it.
5. Safety class and environment: require PLd obstacle detection (ISO 13849-1) and the right machine certification (ISO 3691-4 for CE, ANSI/RIA R15.08 for North America), and confirm floor condition, gradient, temperature range (many indoor AMRs are rated roughly 5 to 40 degrees C), and human traffic level.
6. Power and uptime: match battery chemistry, runtime, and charge ratio to the duty cycle; design opportunity-charging stations against real idle gaps, not just pack capacity, and decide whether spare-battery swapping is ever needed.
7. Fleet management and interoperability: evaluate the fleet-management software (traffic control, route optimization, WMS/MES integration) and require VDA 5050 support if a multi-brand or future-expandable fleet is plausible, to avoid single-vendor lock-in.
8. Total cost of ownership (TCO): sum robot price, software licences (often per robot), chargers and docks, integration, safety validation, training, spare batteries, and annual service. Because an AMR maps itself, first deployment is fast (a 10,000 square-metre map in hours, unboxing to first run in 1 to 3 days), and relocation flexibility usually pays back the premium over fixed-path systems within a few years.

One last and frequently overlooked dimension is vendor serviceability and ecosystem: local spare-part inventory, field service response, software update cadence, the maturity of the fleet manager, and whether the platform is open to third-party integration. Among established suppliers, Mobile Industrial Robots (the MiR250, MiR600, and MiR1350) and OTTO by Rockwell Automation (the OTTO 100, 600, and 1500) publish detailed datasheets and standards compliance and maintain global service networks. ABB, KUKA, and a growing field of specialist and regional makers round out the market for goods-to-person, tugger, and heavy-pallet duties. The right partner is the one whose service footprint and software roadmap still fit your operation in five to ten years, not just the one with the lowest unit price today.