The right fleet size is an engineering result, not a number taken from a robot brochure.
One of the first questions in an AMR project is deceptively simple: How many robots do we need? Suppliers may present a quick estimate based on travel speed and the number of transport requests. That can be useful for an initial discussion, but it is not yet a defensible design. A fleet that looks sufficient on average can fail during shift changes, production peaks, charging windows, or a blocked intersection.
The correct starting point is the operation—not the robot catalogue. Fleet sizing means translating a material-flow requirement into a reliable transport service under real site conditions. The goal is not to maximize robot utilization at any cost. It is to meet the required level of service with acceptable resilience and economics.
1. Map the actual transport demand
Before discussing models or quantities, create a transport matrix. For every flow, record the origin, destination, handling unit, gross load, trips per shift, frequency by hour, priority, pickup/drop-off method, and any time window. Separate scheduled replenishment from event-driven missions and urgent calls. Averages hide the exact moments when the fleet becomes congested.
A useful data set includes at least several representative production days, the busiest periods, and known abnormal conditions. Where reliable historical records do not exist, conduct an on-site observation study, explicitly distinguishing measured data from estimates. Check whether a “transport” means one pallet, two containers, a loaded trolley, or an entire multi-stop mission: changing that definition changes the calculation.
Also ask which transports must be automated. Some short or exceptional moves may be better handled manually. Automating every movement can make the whole system more complex without creating proportional value.
2. Define a complete cycle, not only driving time
A robot's rated speed is rarely its achieved average speed. The relevant unit is mission cycle time, including dispatch, approach, accurate alignment, loading, loaded travel, traffic delays, unloading, confirmation, and the next positioning or empty return. For a forklift AMR, pallet engagement and placement may be more significant than the travel itself.
A practical model can be written as:
Cycle time = dispatch + approach/pickup + loaded movement + traffic/waiting + drop-off + repositioning.
Use route-specific distances and speeds, not one global average. Include doors, lifts, intersections, narrow aisles, crossing priorities and pedestrian areas. If mission times vary considerably, calculate distributions and upper-percentile delays in addition to a mean. The fleet must survive real variation, not merely the best five demonstration runs.
3. Build a transparent first-pass calculation
Suppose a plant requires 20 pallet movements per hour during its busy period. An observed or conservatively modelled cycle takes six minutes on average, including handling and empty repositioning. Demand therefore represents 120 robot-minutes of productive work per hour.
If a robot can realistically dedicate 42 minutes per hour to those missions after allowing for necessary nonproductive time and operational variability, the mathematical first pass is:
120 ÷ 42 = 2.86, rounded up to three robots.
This is an illustrative assumption, not an industry-standard 70% utilization target. Three is not automatically the recommended fleet. It is the minimum result of that simplified model. A fourth unit, route redesign, better dispatching, or different charging arrangements may be justified if a single outage would stop critical production or if demand occurs in short bursts. Conversely, simplifying the handling method may reduce the number needed.
Do not hide the assumptions behind a single supplier-provided fleet figure. Show the transport volume, cycle-time evidence, productive fraction, peak factor, charger impact and chosen redundancy separately.
4. Model shared resources and congestion
Adding robots does not always add equivalent capacity. In a narrow aisle, several AMRs may compete for one traffic section. A single high-use door, lift, pallet stand or staging location can become the actual capacity limit. Mission assignments that send vehicles empty across the site can consume the capacity gained by purchasing another unit.
Evaluate traffic rules, right-of-way priorities, safe separation, queueing, re-routing limits, and the number of physical pickup/drop positions. Consider whether loads are available when missions are released. An AMR waiting for a human-prepared pallet is still occupying fleet capacity even if the delay is caused outside the robot system.
This is where a simple spreadsheet becomes insufficient. For variable flows and interacting resources, a discrete-event simulation or a well-designed operational pilot can expose bottlenecks that average-speed calculations miss. Simulation must be calibrated against observed site data; a visually convincing simulation is not proof of future performance.
5. Treat charging as a fleet resource
Fleet sizing and charging strategy are inseparable. Consider usable battery energy, consumption under the real load, distance, lift cycles, mission duration, charger power, charging efficiency, waiting time, battery aging and reserve requirements. Opportunity charging might reduce battery-related downtime, but only if the schedule creates available windows and the charger is reachable without disrupting traffic.
Ask whether peak flow coincides with the available charging windows. A design with enough vehicles in aggregate may still fail if too many units need energy at the same time. Check what happens after one charger fails and whether the charging point itself creates a traffic conflict.
6. Size for service level and failure recovery
A technically sound proposal states what the fleet must deliver: for example, a defined number of movements within agreed time windows, measured under representative operating conditions. It should also define acceptable queue time, availability assumptions, intervention rate and recovery when a robot is unavailable.
Model at least three scenarios: normal operation, peak operation, and degraded operation with a blocked route or unavailable robot. Calculate whether manual fallback is required. Document which actor owns mission reallocation, blocked-route clearance, escalation and recovery. The most productive theoretical fleet is not always the most reliable operational fleet.
7. Validate the concept before committing to quantity
A credible proof should include representative handling units, realistic floor conditions, actual crossings and a complete mission rather than only point-to-point travel. Confirm loading and unloading tolerances, integration timing and the availability of pallets or trolleys. Record repeatability, travel-time distribution, wait reasons, failed missions and operator interventions.
For formal acceptance, agree on a test script and measurable pass/fail criteria before the tests. Otherwise, teams tend to debate whether the robot “looked fast enough” instead of whether the transport requirement was met.
Fleet-sizing checklist
- Measured origin–destination matrix with hourly and peak demand
- Defined handling unit, load, interfaces and service priorities
- Complete cycle times, including handling, queues and empty travel
- Traffic, door, lift and staging capacity verified
- Charging and battery-aging assumptions documented
- Normal, peak and degraded-operation scenarios evaluated
- Explicit acceptance criteria and fallback ownership agreed
The engineering conclusion
A good AMR fleet is not the largest fleet and not automatically the smallest. It is the fleet that can deliver the required material flow consistently, safely and economically when the site is busy, the floor is imperfect and the operation does not behave according to the brochure. First understand the work. Then calculate the robots.
ČAMINSKI — Engineering Advisory & Project Delivery. For independent fleet-sizing and feasibility review: office@dimitarcaminski.com.