A SCHUNK parallel gripper carries a warranty of “24 months or 5 million cycles.” Two clocks, and whichever runs out first ends the cover. Robot hands are sold on that structure, and once you see it the economics stop being about durability and become about which clock binds. Run the hand hard and the counter binds. Run it lightly and the calendar binds, and you pay for cycles you never used. There is no throughput setting at which the hand stops being a scheduled cost.

The only published maintenance table is for the simplest hand

Start with the least ambitious hand in the catalogue: a two-finger gripper with a single actuator. Robotiq publishes a maintenance schedule for its 2F-85 and 2F-140, and it runs on both clocks at once. Service is due after specified usage, “measured in time (normal 40h week) or in cycles.” The cycle columns carry periodic inspection at 1 million cycles, fingertip replacement at 1 million, and an overhaul at 2 million. The manual defines the unit it is counting: “A cycle is defined as a go to requested position command that results in grip force being applied.” Roughly, one grasp.

Two details matter more than the numbers.

The first is that the fingertips are consumables, with a scheduled replacement point and a condition that overrides it: “Replace fingertip before if wear is visible.” The second is who pays for the overhaul. The manual is explicit: it “is recommended after 2M cycles and is done by Robotiq at the user’s expense.” The vendor decided that and wrote it down.

Note where this lives. The specifications chapter gives grip force, finger speed and repeatability, no cycle rating. The wear schedule sits in the maintenance chapter, the part buyers read after they own it.

What a shift does to that number

A cycle count becomes a date only once you know the rate, so take a measured one. Team Delft won the Amazon Picking Challenge in 2016 and measured their robot against the human benchmark: “Speed is still far from human performance ( ~ 100 items an hour, compared to 400 items an hour in the case of a human).”

Take the slower figure first, the one least helpful to my argument. At 100 picks an hour, 1 million cycles is 10,000 operating hours. On a single shift, eight hours a day and 250 days a year, that is 2,000 hours a year, which is the 40-hour week Robotiq calls normal. So the first fingertip replacement lands after five years and the paid overhaul after ten. On those numbers there is no story here.

Now the other figure, because robots reaching the human rate is why anyone is funding this. At 400 picks an hour, 1 million cycles is 2,500 hours, or fifteen months on that same shift. The overhaul arrives at two and a half years. Run two shifts and halve both.

Nothing about the hand changed between those two paragraphs. What changed is the number the bull case rests on. On the counter alone, as I read it, every gain in picks per hour is spent twice: once as revenue, once as wear.

One assumption is doing work there: I am treating one pick as one cycle. Robotiq counts a cycle when grip force is applied, so a regrasp or a failed attempt is another one. That pushes the intervals shorter, not longer.

The current manual does not say what the time half of that interval is. An earlier revision does. It heads the same two columns “Semiannually (or 1 M cycles)” and “Annually (or 2 M cycles),” whichever comes first, the structure of the SCHUNK warranty. On a single shift the crossover sits at about 1,000 picks an hour. Below that the calendar binds and you change the pads twice a year whether you ran 800,000 grasps or 200,000. Both of Team Delft’s rates are below it.

So the fast operator is billed by the counter and the slow one by the calendar, and neither gets a hand that is free when it is idle.

Twenty motors, and no number at all

Now scale the mechanism count. The Shadow Dexterous Hand is built for dexterity rather than gripping: “The thumb has 5 degrees of freedom and 5 joints. Each finger has 3 degrees of freedom and 4 joints,” driven by twenty Smart Motor nodes packed into the hand base, each pulling a pair of tendons. Against one actuator and two fingers, that is twenty actuators, twenty gearboxes, and forty tendons.

Its technical specification states no maintenance interval, no cycle rating and no service life. I checked it for all three. That is not a claim that no figure exists, only that the downloadable document does not carry one while a two-finger gripper’s manual does.

So the one hand that publishes a wear schedule is the one with the fewest parts to wear, and the figure goes missing exactly where mechanism count says it is hardest to hold. That is the shape of the joint in the last post, and a fourth entry for the numbers demos do not give you.

What it costs

A payback calculation for a picking robot scales revenue with picks per hour. Check what it does with maintenance. If that line is a fixed percentage of capital cost per year, above the crossover the two cannot both be right. Maintenance on a cycle-rated part accrues per cycle, so raising throughput raises the charge in proportion. Holding it flat while revenue climbs flatters exactly the aggressive scenarios that justify the purchase.

Below the crossover the fixed charge is correct, and the problem moves to the other side of the ratio. A fleet picking at Team Delft’s 100 an hour carries the same hand bill as one picking four times faster, spread over a quarter of the output. Either way, the hand bill and the output it supports run on different clocks, and a model has to say which one it uses.

The second cost the two-finger gripper has priced and the dexterous hand has not. Robotiq says the overhaul is at the user’s expense. Where no schedule is published, that allocation gets settled later, per contract, by whoever has more leverage. An undecided cost is not a small one. It is an unpriced one, and my expectation is that unpriced maintenance liabilities land on the party that already took delivery.

Where I could be wrong

Fewer actuators may be the answer. The i-HY hand is “a medium-complexity manipulator with 5 actuators” whose authors argue that “through strategic use of compliance and underactuation, the hand is able to manipulate objects gently, and is also mechanically robust.” If production hands converge on that rather than twenty-motor anthropomorphism, mechanism count stops growing and this argument weakens.

These may be conservative numbers, not failure points. SCHUNK’s 5 million is a warranty limit and Robotiq’s 2 million is a recommendation. Neither is a measured mean time to failure. Real hands may run well past both, and a warranty limit is set by finance as much as by a test rig.

The schedule is explicitly not universal. Robotiq writes that the intervals are for “the average normal usage of the gripper” and “must be adjusted according to environmental conditions”, listing temperature, humidity and debris. A clean, gentle application may run well past them, and my arithmetic assumes an average one.

Publication would settle it. If makers of multi-fingered hands published cycle ratings the way gripper makers do, none of this inference would be needed. That is the disclosure to watch for.

Until it appears, every payback model for a dexterous hand carries a maintenance line nobody has priced, on a part no vendor has agreed to pay for. The buyer holds it either way.