A carton picking robot carries far less weight than a pallet AGV, yet its wheels face a stricter specification. Total vehicle weight is typically 150 to 300 kg, which works out to 30 to 75 kg per wheel, against 200 to 800 kg for a pallet mover. Payload is not the problem. Repeatability is.
The reason is the task. A pallet AGV delivers to a floor position and tolerates 20 to 50 mm of error. A carton picking robot reaches into a shelf compartment at the seventh level and extracts one tote, with a docking tolerance usually held to plus or minus 10 mm. Once the machine has to place rather than simply arrive, the wheel stops being a wear part and starts behaving like a precision component.
The arithmetic is unforgiving. Take two drive wheels of 150 mm diameter on a 500 mm wheelbase. If their diameters differ by 0.5 mm, the vehicle yaws 0.38 degrees after one metre of travel and 1.15 degrees after three metres, which translates to roughly 30 mm of lateral drift at the end of that run. Against a docking window of plus or minus 10 mm, a pair of wheels that looks identical is enough to make the robot stop off-target.
This guide sets out how to specify a carton picking drive wheel: the pairing tolerance that follows from the docking accuracy, the hardness choice that follows from floor and duty cycle, and the wear threshold that should trigger replacement. HANKE supplies both hardness classes discussed here, HANKE's Eamflex 93A high-wear compound and the Saxflex 75A floor-protection compound, and inspects wheel geometry on a CHOTEST coordinate measuring platform at 1 micron resolution.
Three rules summarise the practical conclusion. Pairing tolerance should be held to 0.1 mm or better. Hardness is selected on two axes, floor finish and duty cycle, not on wear resistance alone. Replacement should be triggered by measured diameter loss rather than by calendar time.
Parameter | Typical Range | Notes |
Load per wheel | 30 to 75 kg | 150 to 300 kg vehicle, four or six wheels |
Travel speed | 1.0 to 2.0 m/s | Slower than a high-speed transport AGV, faster than line conveying |
Start-stop cycles per day | 300 to 2,000 | Set by shift pattern and slot density; the main fatigue driver |
Drive wheel diameter | 100 to 150 mm | Set by chassis height and obstacle clearance |
Left-right diameter pairing | 0.1 mm or less | Deviation shows up directly as straight-line drift |
Shore hardness | 93A or 75A | Wear priority versus floor protection priority |
Diameter loss, inspection point | 0.3 to 0.5 mm | Measured, not estimated from service time |
Diameter loss, replacement point | 0.8 mm | Replace as a pair, never one side only |
Retained friction coefficient | 0.6 or above | Below this, slip probability rises sharply |
Docking accuracy requirement | plus or minus 5 to 10 mm | Set by slot geometry and the pick mechanism |
Of these ten parameters, only three actually decide whether the wheel works: the pairing tolerance, the hardness, and the retained friction coefficient. The rest are inputs to the selection, not outputs of it.
Two mobile robots can look almost identical at the chassis and still demand completely different wheels. The difference is not in the engineering, it is in the job.
A pallet AGV moves a load from point A to point B, and point B usually allows some slack. A carton picking robot reaches into a specific compartment, at a specific level, and pulls out a specific tote. The compartment is only marginally larger than the tote, and the pick mechanism has to line up before it extends. Change the task and the role of the wheel changes with it.
Dimension | Carton Picking Robot | Pallet Transport AGV | Production Line Conveyor |
Load per wheel | 30 to 75 kg | 200 to 800 kg | Varies widely by line design |
Accuracy requirement | 5 to 10 mm docking | 20 to 50 mm arrival | Fixed position; wheels carry no locating duty |
Start-stop cycles per day | 300 to 2,000 | 50 to 200 | Continuous running, few stops |
Motion profile | Short runs, high frequency, hard stops | Medium to long runs, mostly steady state | Continuous rolling |
Floor condition | Epoxy coating, some PVC | Concrete, epoxy, steel plate | Steel roller or belt |
Dominant failure mode | Position drift from diameter loss | Tread worn through or debonding | Roller surface damage, slip |
Replacement trigger | Measured diameter loss | Visual tread condition | Roller surface condition |
The last two rows carry the point. A carton picking wheel is rarely replaced because it has worn out. It is replaced because it has worn enough to change dimension while still looking serviceable. That single distinction separates this application from the other two.
The first is a wheel that looks fine but no longer works. The tread pattern is intact, there is no visible damage, and the maintenance team judges it usable. Meanwhile the pick success rate has fallen from 99 percent to 85 percent. Fault-finding usually starts with the vision system, moves to the fleet scheduler, and only reaches the wheels several days later.
The second is drift caused by mixing an old wheel with a new one. In practice, only the worn wheel gets replaced. The two diameters then differ by a fraction of a millimetre and the robot begins tracing a gentle arc. This is frequently misread as navigation parameter drift, which leads to repeated tuning and makes the problem worse.
The third is a calibration interval set by guesswork. Most sites calibrate quarterly regardless of conditions. But diameter loss depends on floor roughness, cycle frequency and load, and can vary by a factor of two between workshops. A fixed calendar either wastes labour on unnecessary calibration or leaves a drifted robot uncorrected.
There are exactly three paths by which a wheel affects positioning: a diameter mismatch producing heading error, diameter loss producing odometry error, and unsuitable hardness producing compression at the stop. All three can be calculated from geometry, and once calculated they indicate where the tolerances need to sit.
A differential-drive robot travelling in a straight line turns both drive wheels at the same rate. If the two wheels have different diameters, the same rotation produces different travel distances, and the vehicle is in fact tracing a very large arc.
The heading angle can be estimated with a simple relationship: the tangent of the heading error equals the relative diameter difference multiplied by travel distance and divided by wheelbase. For a 150 mm drive wheel on a 500 mm wheelbase, a 0.5 mm diameter difference produces 0.38 degrees of yaw after one metre and 1.15 degrees after three metres. Using the circular-arc approximation, the resulting lateral offset is the travel distance multiplied by half the heading angle, which comes to roughly 30 mm over three metres.
That 30 mm figure matters because carton picking robots generally work to a docking tolerance of plus or minus 10 mm. A half-millimetre diameter difference, traversed over a three-metre gap between floor markers, is enough to push the error to three times the tolerance.
Diameter Difference | 1 m Travel | 3 m Travel | 5 m Travel | Practical Meaning |
0.1 mm | 0.7 mm | 6.0 mm | 16.7 mm | Controllable at 3 m marker spacing |
0.3 mm | 2.0 mm | 18 mm | 50 mm | Outside a plus or minus 10 mm tolerance |
0.5 mm | 3.3 mm | 30 mm | 83 mm | Docking failures become likely |
Figures assume a 150 mm diameter, a 500 mm wheelbase and the circular-arc approximation. Real values will shift with slip and floor flatness. The conclusion does not: pairing tolerance has to be held at the 0.1 mm level, and correcting it after installation is the more expensive route by a wide margin.
This also explains an otherwise puzzling design choice. Carton picking robots almost always run on dense floor markers, spaced one to three metres apart. That is not purely a navigation decision. Part of it is forced by the physical behaviour of the wheels.
A closed-loop system resets its pose error each time it passes a marker, so the cumulative error does not grow without limit. HANKE's field experience with AGV and warehouse robot wheels points the same way. But diameter loss affects more than accumulation. It also changes how far the odometry can be trusted between corrections.
Every 0.1 mm worn from the diameter changes the rolling radius by about 0.067 percent. Over one metre the odometer reads 0.67 mm away from the true distance; over ten metres, 6.7 mm. That magnitude is not large on its own. The difficulty is that it combines with slip error, and the two act in opposite directions. A reduced diameter makes the reading high, slip makes it low, and each masks the other, which is why the two are so hard to separate during diagnosis.
The greater difficulty is that the two wheels do not wear at the same rate. The outer wheel travels further in every turn and therefore wears faster. Left-right diameter difference grows with service, which means a pair that met specification at dispatch may no longer meet it halfway through its life.
Polyurethane is an elastomer and deforms under load. The contact patch flattens, and the effective rolling radius becomes smaller than the free radius. The amount of compression depends on hardness and load: lower hardness and higher load both increase it.
The effect on positioning appears at the stop. When the robot brakes hard, the wheel must absorb kinetic energy through deformation and then recover its shape, and during that interval the wheel centre moves slightly relative to the floor. Starting is the mirror image. Where the two sides compress differently, because one is harder or carries more load, each start and stop introduces a lateral displacement that is difficult to predict.
Hardness is therefore not only a wear parameter. It influences life and accuracy at the same time, which is why the same robot model running on epoxy in one facility and worn concrete in another should not share a wheel specification.
A parameter is specified correctly when it can be traced back to a usable tolerance boundary. The four below are ordered by importance.
Docking Accuracy | Pairing Tolerance | Single Wheel Tolerance | Note |
plus or minus 5 mm | 0.05 mm or less | 0.1 mm or less | Full CMM inspection; sampling is not sufficient |
plus or minus 10 mm | 0.1 mm or less | 0.2 mm or less | Common case; supply as matched pairs |
plus or minus 20 mm | 0.2 mm or less | 0.3 mm or less | Batch sampling acceptable |
Pairing tolerance and single-wheel tolerance are two different things. Two wheels can each sit inside their individual tolerance while still being one at the upper limit and one at the lower, which produces a difference that fails the pairing requirement. Pairing therefore has to happen after individual inspection, and that in turn requires a supplier able to ship matched pairs rather than loose pieces in a carton.
Hardness selection for carton picking wheels follows a two-dimensional rule: floor flatness sets the lower bound, cycle frequency sets the upper bound.
On flat, hard floors such as epoxy or PVC, contact stress is concentrated and the wheel needs higher hardness to resist wear, which places the specification around the 93A class. Where the floor has joints, local unevenness, or a coating that needs protection, a 75A compound provides a larger contact patch and better deformation cushioning.
HANKE builds these two classes as distinct tread systems rather than one compound adjusted up or down. HANKE's Eamflex 93A is specified for wear resistance and load carrying on hard, high-frequency floors. HANKE's Saxflex 75A is specified for floor protection, quiet running and clean environments. The two differ in chemical system and crosslink density, so they also differ in compression set, which is the property that governs how much the wheel deforms at a hard stop.
Wear resistance is specified reliably by asking for the method and the number. The established method is DIN 53516, which measures volumetric abrasion loss against a standard abrasive surface. Lower values indicate greater wear resistance. The advantage of the method is that it is reproducible and comparable, and it does not depend on how a supplier chooses to describe the material.
One caveat applies. DIN 53516 measures the wear behaviour of the material, not the service life of the wheel. Real life also depends on floor roughness, load and the degree of slip. The number is therefore useful for comparing compounds and unreliable for predicting how many months a wheel will last. Suppliers who convert a DIN figure directly into an expected service life are extrapolating well beyond what the test supports. HANKE provides DIN 53516 values for its tread compounds on request, stated as measured rather than as a service life estimate.
Polyurethane generally offers higher friction than rubber on clean, dry floors, but the value changes with surface condition. Oil film, standing water and dust all reduce friction. The wheel slips, odometry readings diverge from actual displacement, and positioning fails almost immediately.
The check is direct. After a period of service, measure friction on the actual travel path with a portable tester. Below 0.6, either the floor cleaning regime or the compound needs attention. Positioning faults are usually blamed on the algorithm first, yet a clean floor frequently resolves the problem, which suggests the sequence of investigation should start at the floor rather than the code.
The sequence matters. Each step consumes the output of the one before it, so the order cannot be rearranged.
Work backwards from the docking requirement. If the docking tolerance is plus or minus 10 mm, vision error, mechanism clearance and floor flatness each consume part of it. The share available to wheel-induced deviation is typically 3 to 5 mm. Inverting the geometry, a 3 mm lateral allowance over a 3 m marker gap corresponds to a diameter difference of roughly 0.05 mm. Only after this step is the required tolerance known.
Set the pairing tolerance from Step 1 and specify the supply method at the same time. Where the pairing tolerance is 0.1 mm or tighter, require matched pairs with matched identification. Buying loose pieces and pairing them on site rarely works, because the site usually lacks the metrology to do it properly. HANKE ships matched pairs with the measured diameters recorded, so the pairing is verifiable rather than assumed.
Flat hard floors with frequent cycling call for the 93A wear-resistant class. Clean, quiet environments where the floor must be protected call for the 75A class. Where the two requirements conflict, for example a cleanroom installation that also runs high cycle counts, protect the quiet and clean requirement and recover life by increasing the wheel diameter instead.
Do not set the calibration interval by calendar. Set it by measured diameter. Record the initial diameter when a wheel enters service, then measure every two weeks with a caliper or laser gauge, averaging three points. At 0.3 to 0.5 mm of loss, halve the calibration interval. At 0.8 mm, replace the pair.
Diameter Loss | Status | Recommended Action |
0 to 0.3 mm | Normal | Keep the existing calibration interval |
0.3 to 0.5 mm | Entering the wear phase | Halve the calibration interval, increase monitoring |
0.5 to 0.8 mm | Deviation risk zone | Check docking success rate, prepare spares |
Above 0.8 mm | Replace | Replace as a pair, recalibrate afterwards |
The value of this table is that it converts a judgement into a measurement. Maintenance crews working from experience alone tend to apply different thresholds on different shifts, which is precisely what a written criterion prevents.
Q1: Can a carton picking robot use the same wheels as a standard AGV?
Not directly. A standard AGV wheel is designed primarily around load carrying and wear, with dimensional tolerance usually controlled at the millimetre level. A carton picking robot needs 0.1 mm diameter consistency plus matched pairing. The compounds may be similar, but the inspection and pairing stages are not, and a direct substitution typically produces position drift within one to two weeks of service.
Q2: The wheel looks undamaged. Why replace it?
Because the replacement criterion is dimensional, not visual. Positioning depends on consistent diameter, so a wheel with an intact tread but 0.8 mm of diameter loss is visually acceptable and functionally out of specification. The practical answer is a diameter log with a measurement every two weeks, so the decision is made on data.
Q3: Can only one side be replaced?
It is not advisable. A new wheel and a used one typically differ by 0.3 mm or more in diameter, and the robot will begin to drift as soon as the replacement is fitted. Replace as a pair, keep the removed wheel as an emergency spare, and mark its measured diameter on it so it can only be paired with a wheel of matching size.
Q4: How do I choose between 93A and 75A?
Two variables decide it: floor and duty cycle. Flat hard floors such as epoxy or PVC with frequent stops call for the 93A wear-resistant class. Where the floor needs protection, or the environment has clean or quiet requirements, choose the 75A class. In the rare case where both conditions apply, protect the clean and quiet requirement first and compensate for life by increasing wheel diameter.
Q5: How much diameter loss is acceptable?
Two thresholds are practical. At 0.3 to 0.5 mm the wheel enters the inspection phase and the calibration interval should be halved. Above 0.8 mm, replace the pair. For installations working to plus or minus 5 mm docking accuracy, tighten the replacement threshold to 0.5 mm.
Q6: Why not set a fixed quarterly calibration interval?
Because wheel wear is not linear in time. Floor roughness, cycle frequency and load all affect the rate, and the same robot model can wear twice as fast in one workshop as in another. A calendar-based interval either wastes labour or leaves a drifted robot uncorrected. Interval should follow measured diameter.
Q7: Polyurethane or rubber for a carton picking robot?
Polyurethane generally offers better load capacity, wear resistance and dimensional stability, with smaller deformation, which suits applications where diameter must remain consistent. Rubber offers better elasticity, damping and noise reduction. Because carton picking is driven by positioning accuracy, polyurethane suits most installations, while noise-sensitive clean or medical environments may benefit from a lower-hardness polyurethane compound as a compromise.
Q8: How can I tell whether a supplier's dimensional consistency is reliable?
Ask three questions. Is there a measurement record for every wheel, and is it caliper or CMM based. Can the supplier ship matched pairs, and where is the pairing tolerance written down. What test method produced the wear figures, and can the measured values be provided. A supplier who answers all three specifically is usually consistent; one who answers in generalities is not.
The wheel requirement for carton picking robots comes down to treating a component that is normally a consumable as a precision part. Load capacity and wear resistance still matter, but they are the entry requirement rather than the differentiator. What decides whether the machine works is diameter consistency and how predictably that diameter degrades.
Three sentences capture the method. Hold pairing tolerance to 0.1 mm or better. Select hardness on floor and duty cycle together, not on wear alone. Trigger replacement on measured diameter rather than on service time. Applied together, these three eliminate the majority of position drift faults.
For equipment builders, wheel selection is worth involving early, at the design stage, rather than discovering at commissioning that docking is unstable. The wheel is a small part of the bill of materials. The after-sales cost it can generate is not. That is the perspective HANKE works from when specifying a wheel for a robot platform.