Forty minutes a day: what the world's largest robot fleet publishes about its own utilisation

The release of 6 August 2026 gives three numbers that are large on their own: more than 50,000 deployed robots, more than 5.3 million autonomous operating hours in the first half, 3.7 billion square feet cleaned, plus 68 percent year-on-year growth in the fleet and 24 percent more autonomous hours.

The number that is not in it follows from the first two. 5.3 million hours divided by 50,000 robots is 106 hours per robot in the half; divided by 181 days that is 0.59 hours a day — around 35 minutes.

That is a lower bound, and it is fair to say so. The 50,000 is the figure at the end of the half; at the start there were fewer, so the hours are spread across a smaller average fleet. Spread the annual growth of 68 percent evenly across the year and the fleet stood at around 40,000 in January and averaged around 45,000 over the half — then it is 0.65 hours, around 40 minutes. The fleet also comprises not only scrubbers but other device types, and what exactly counts as an autonomous hour is the supplier's definition. Between 35 and 40 minutes per robot per day, then, is where the value sits, with all the reservations such a calculation has to carry.

It is nevertheless the most honest figure on the utilisation of autonomous cleaning robots that is publicly available, because it comes not from a case study but from the sum across a fleet whose supplier would have good reasons to show a larger number.

That forty minutes is little does not have to be asserted. It can be read off the supplier's own yardstick. On 4 May 2022 Brain Corp published, in a guide to fleet management, examples of the KPIs an operator should use to steer its robots:

• Autonomous use should be at 80 percent or above, measured against manual operation of the same machine.
• Each robot should have a daily target for hours of use, “such as two hours per day” — two hours a day.
• Each robot should reach a daily area, “such as 30,000 square feet” — around 2,800 square metres.
• The number of assists, meaning interventions by staff, is the indicator for maintenance needs and retraining.

Between the example target of 2022 and the fleet average of 2026 lies a factor of three. That is not a statement about the quality of the machines; it is a statement about how much of the theoretically available time a robot actually works autonomously in regular operation once thousands of operators have put it in their buildings. Anyone who wants to know after month three where they stand now has both ends of the scale: the target the supplier itself proposes, and the value its fleet reaches.

And the metric that lies between them is not availability. It is autonomous hours per robot per day, against a target set in-house. Why availability cannot see this gap we set out in progress detection as an operating metric; here the subject is the number that replaces it.

The explanation for the factor of three is not in the press release but in the technical data of the machines that make up such fleets. Kärcher publishes a datasheet for its autonomous scrubber KIRA B 50, dated 28 August 2024. Two lines from it suffice.

Theoretical area performance, autonomous: max. 2,365 m²/h. That is the number in every quotation.

Area performance per tank fill: approx. 1,830 m². That is the number that sets the day.

The fresh-water tank holds 55 litres, the dirty-water tank the same. At theoretical rated performance, 1,830 square metres are cleaned after 46 minutes — and the tank is empty. The battery, according to the datasheet, lasts around 3.5 hours. The machine could run a good four tank fills before it has to charge, and it runs exactly one if nobody refills it.

That is the mechanism behind the fleet figure. A robot that drives autonomously but does not get water autonomously works for as long as one tank lasts and then waits. The waiting time is green on the availability dashboard, zero in the metric “autonomous hours per day”, and on the shift plan it is a person walking to the machine. At 46 minutes per tank and a two-hour daily target that is three trips per robot per day; with ten robots, thirty.

Kärcher offers a docking station for the KIRA, and its description in the accessories part of the datasheet explains why: it “reduces manual interventions to a minimum and extends the range and area performance of the robots”. That is, word for word, the metric from the previous section, turned into a component. According to the manufacturer the station refills fresh water, empties and rinses the dirty-water tank and charges the battery; it is wall-mounted in the immediate vicinity of a fresh-water connection, a power socket and a floor drain, with two by two metres of clear space in front. That is building services, not robotics, and it is as a rule the part that is not finished at go-live. How charging a fleet can be organised is set out in the article on charging infrastructure; water is the harder half of the same question.

The first weeks of a deployment are measured by availability and completed jobs, and rightly so, because there the question is whether the machine works in the building. From month three the question is whether the building uses the machine. Four numbers suffice, all from data any fleet software keeps anyway.

1. Autonomous hours per robot per day, against target. The target is set by the operator, not the manufacturer; two hours is a starting point the supplier itself proposes. Below one hour the machine is a pilot that was never closed.
2. Autonomous share of total runtime. A scrubber that is mostly pushed by hand is an expensive scrubber. Brain Corp's example value is 80 percent.
3. Assists per autonomous hour. Not the number of interventions but the number per hour, otherwise the metric rewards the robot that does not run. If it rises, the site has changed — new displays, new pallets, a door — not the machine.
4. Tank fills per shift, with who did it. Station or person. This number explains the first: where a person fills, the ceiling on autonomous hours is the number of their trips.

Together the four numbers yield a statement an availability figure never delivers: whether a second robot would bring more autonomous hours or only a second tank waiting for someone.

The effect of the metric is organisational, and it comes quickly. As soon as autonomous hours per robot per day sit next to the target, it becomes visible that the machine is not idle because of the machine.

It is idle because the time window is missing: the hall is occupied by day, the night shift has other duties, and nobody scheduled the 10:30 p.m. start. It is idle because the water is missing: the tank is empty, the station not built, the connection at the other end of the building. It is idle because the route is missing: an area was rearranged, the old route breaks off, and the person who re-records it is not on shift.

Each of these causes has an owner, and none of them is the manufacturer. That is the real function of the metric after month three: it moves the conversation from “is the robot running” to “who made sure today that it could”. A pilot does not answer these questions, because during its weeks the manufacturer refills the water and re-records the route — which is why it proves so little. The utilisation figure is what ends the pilot.

Three caveats, because the arithmetic is only as robust as its inputs.

The forty minutes are derived, not reported. Brain Corp publishes no utilisation per robot. The figure follows from two reported quantities and an assumption about growth within the half; it describes an order of magnitude, not a measurement. The fleet contains different device types, and the definition of an autonomous hour is the supplier's.

The datasheet is one model from one manufacturer. 2,365 m²/h and 1,830 m² per tank apply to the KIRA B 50 as of August 2024. Other machines have other tanks, other working widths and other ratios of water to battery. What transfers is the structure — tank against battery, station against person — not the minute count. Anyone doing the arithmetic for their own machine needs exactly those two lines from their own datasheet.

Two hours is an example, not a benchmark. The 2022 guide names the two hours and the 30,000 square feet explicitly as examples of targets an operator sets. A logistics centre with 20,000 square metres of floor and a supermarket with 2,000 have different targets; the point is that there is one and that it sits next to the actual figure.