Repeatability vs Accuracy: The Robot Spec Buyers Confuse Most

Repeatability is the spread when a robot returns to a taught point, typically ±0.02 mm to ±0.1 mm. Accuracy is the error when it moves to a calculated coordinate, and on an uncalibrated arm that is commonly 0.5 mm to 3 mm, ten to fifty times worse. Datasheets print repeatability because it is the flattering number, and offline programming is the moment that difference stops being academic.

The two definitions, precisely

Both are defined in ISO 9283, which also fixes the test conditions, and that is why comparing figures from different conditions is meaningless.

• Pose repeatability is the closeness of agreement between attained poses after repeated visits to the same commanded pose, measured over 30 cycles at a defined test pose in a defined cube, at rated payload and rated speed. It captures random error: encoder noise, backlash reversal, thermal short-term drift.
• Pose accuracy is the deviation between a commanded pose and the mean of the attained poses. It captures systematic error: link length tolerances, joint offsets, gearbox eccentricity, deflection under load, and the difference between the nominal and actual tool centre point.

An arm can be extremely repeatable and badly inaccurate. It goes to the same wrong place every time, which is exactly what makes teach-and-repeat programming work despite poor accuracy.

When each one governs

The fifth row is the one that catches production engineers. Two robots of the same model, same batch, same firmware, will not run the same program to the same result, because their systematic errors differ. Without calibration, moving a program between arms requires re-touching points, and on a 300-point program that is a day of work per machine.

How accuracy is actually recovered

Absolute accuracy is a solvable problem, in three escalating steps.

1. Tool centre point calibration. A four-point or six-point procedure on the pendant, 10 to 30 minutes. Removes the largest single contributor on most cells, which is a tool frame that was estimated rather than measured.
2. Base and workpiece frame calibration. Ties the robot to the fixture geometry. Another 20 to 60 minutes, and it is what allows a fixture to be moved and the program to survive.
3. Kinematic calibration. Measuring the arm with a laser tracker or a photogrammetric system across dozens of poses, then fitting corrected link parameters into the controller. Typically half a day of measurement and it brings absolute accuracy to roughly 0.1 mm to 0.3 mm on a mid-size arm.

Many manufacturers sell step three as an option, sometimes described as absolute accuracy calibration. For any cell running offline-generated paths it usually pays for itself in avoided touch-up work on the first program.

Reading a datasheet without being misled

• Check whether the quoted figure is repeatability or accuracy. If the word accuracy does not appear, the arm is very likely not specified for it.
• Check the test conditions. ISO 9283 requires rated payload and rated speed at a defined pose; a figure measured unloaded at low speed is a different number.
• Look for the sign convention. A stated ±0.03 mm is a different claim from a stated 0.03 mm range.
• Remember temperature. Cold-start drift of 0.05 mm to 0.2 mm over the first 30 to 60 minutes of operation is normal, and it is not covered by either figure.
• Repeatability degrades with wear. Gearbox backlash after several years of duty shows up here first.

Frequently asked questions

Sources