Accuracy, Repeatability and Resolution Are Not the Same Number
Three specifications on every positioner datasheet get used as though they were interchangeable. They are not, and the difference decides whether a system points where you told it to.
Ask three suppliers for a positioner accurate to a hundredth of a degree and you will get three quotes, all compliant, differing by a factor of four in price. The specifications are not lying. They are answering different questions, and the datasheet rarely says which.
This paper separates the three figures that get confused most often, shows what each one is measured against, and works through what happens to them once a payload, a temperature range and a gearbox are involved.
Three questions, three answers
Resolution is the smallest change in position the system can report. It comes from the feedback device: a 22-bit encoder divides a revolution into 4,194,304 parts, so it can resolve about 0.00009°. That is a property of the counter, not of the machine it is bolted to.
Repeatability is how closely the axis returns to the same place when you send it there again. It says nothing about whether that place is correct — only that the error is consistent.
Accuracy is how far the axis actually ends up from where you asked. It includes every error the other two figures ignore.
The middle target is the one that catches buyers out. A tight group is easy to demonstrate and says nothing about where the group sits.
The middle case is worth dwelling on, because it is the one a supplier can demonstrate cheaply. A positioner with a systematic offset — a mis-set index, an encoder mounted eccentric to the shaft — will repeat beautifully and point wrongly every time. If the offset is stable it can be calibrated out. If it drifts with temperature, it cannot.
Resolution you can command and resolution you can reach
A control system will accept a command for one encoder count. Whether the axis moves is a separate question.
Below some threshold, friction in the bearings, the gear mesh and the shaft seals absorbs the commanded step entirely. The motor develops torque, the structure winds up elastically, and nothing rotates until the commanded angle is large enough to break stiction. Then it goes — usually further than asked. The datasheet number was real; the motion was not.
This is why a resolution figure means nothing without a repeatability figure beside it. Resolution finer than repeatability is arithmetic, not capability. On our own units the two are quoted together for exactly this reason:
| Model | Axis resolution | Positioning accuracy |
|---|---|---|
| PED08 | 0.02° | 0.12° |
| PED10 | 0.002° | 0.028–0.032° |
| PED12 | 0.002° | 0.028–0.038° |
| PES10 | — | 0.028–0.042° |
| PEG03 | 0.0056° | 0.056° |
Notice the ratio. Resolution is roughly an order of magnitude finer than accuracy on every line. That is the honest relationship: the encoder must resolve considerably better than the mechanism performs, or the feedback loop cannot see the error it is meant to correct. A datasheet where the two numbers are close is describing a system that cannot use its own encoder.
One direction or two
Repeatability quoted for approaches from a single direction hides backlash. Come back to the same target from the other side and the axis stops short by the play in the drive train.
For a surveillance positioner this matters more than it first appears, because scanning reverses direction at the end of every sweep. A unit with 0.01° unidirectional repeatability and 0.15° of backlash will return to each end of its scan 0.15° from where it was on the previous pass — a wobble that looks like drift on the operator’s screen and is nothing of the kind.
Our electromechanical positioners are specified at zero backlash, which is a statement about the drive architecture rather than a tolerance: the mesh is preloaded so that the flanks never separate. It is the reason a scan can be reversed without a settling allowance.
What a payload does to all three
Every figure above is measured on a bench with a defined load. A real system adds three things.
Six degrees of freedom, two of them driven. A pan axis rotates in yaw. The other five — pitch, roll and three translations — are supposed to stay at zero, and are held there by bearings with finite stiffness. A payload mounted with its centre of mass away from the bearing axis converts its own weight into a moment that tilts the axis slightly. The encoder, which reads shaft angle, sees nothing wrong.
Structural compliance. The positioner is at the top of a mast or a tripod, and neither is infinitely stiff. Wind load on the payload deflects the whole assembly. The line of sight moves; the encoder still reads the commanded angle.
Temperature. Aluminium moves about 23 µm per metre per kelvin. Across a −32 °C to +60 °C operating range, a structure that is dimensionally perfect at 20 °C is not perfect at either end, and an encoder mounted on one material and read by a head mounted on another will drift with the difference.
Proportions are illustrative and depend entirely on the installation. The ordering is not: on a masted system the structure usually dominates, and buying a finer positioner changes almost nothing until it is fixed.
The practical consequence is that the positioner specification is a ceiling, not a prediction. It tells you what the unit contributes. What the system delivers is that figure combined with everything it is bolted to.
How to ask for a number you can hold someone to
Four qualifiers turn a marketing figure into a specification:
- Per axis or for the system? A two-axis figure quoted without a per-axis qualifier is ambiguous by a factor of √2 at best.
- At what load? No-load accuracy is a bench measurement. Ask for it at the payload mass and centre-of-gravity offset you intend to fly.
- Over what temperature? A figure valid at 20 °C says nothing about a winter deployment.
- Unidirectional or bidirectional? If it is not stated, assume unidirectional and ask what the backlash is.
A supplier who can answer all four has measured the unit. A supplier who answers with one number has quoted the encoder.
Where this leads
Once the terms are clear, the next question is what the numbers mean for a target on the ground rather than an angle on a shaft. Angular error scales with range: the same 0.03° that looks negligible on a datasheet is half a metre at 1 km. Building that into a full system error budget — sensor, positioner and structure together — is the subject of a separate paper on target location error.
Technical Discussion and Quotation
Tell us the platform and the constraint. Our engineering team answers with a configuration that fits it, usually within two working days.
