About RepeatabilityFrom Wikipedia:
Accuracy and repeatability are different measures. Repeatability is usually the most important criterion for a robot. ISO 9283 sets out a method whereby both accuracy and repeatability can be measured. Typically a robot is sent to a taught position a number of times and the error is measured at each return to the position after visiting 4 other positions. Repeatability is then quantified using the standard deviation of those samples in all three dimensions.
ST Robotics uses a modified form of the ISO 9283 test with less motion between samples. Positions are taught that test X Y and Z dimensions. Each dimension is tested individually. The robot goes into position against an LVDT (electronic dial gauge) and readings are taken automatically by the controller via the analog I/O card.
A standard test is conducted over a 24 hour period with a short test over 5 hour period. Within the test the position is measured by the LVDT 30 times at 1 minute intervals. The robot then calibrates and another 30 samples are taken over the half hour period. From the readings we get it can be seen that calibration itself introduces some variation or lack of repeatability. For applications requiring much higher repeatability we advise absolute calibration using what we call a nest. Instructions for doing this are in our manuals.
This video shows one cycle of the test for the Y dimension on an R12. The gauge is at Y=350mm
Note that any error in pitch also contributes to the drift. The graph below shows the a typical result from the above test. Vertical axis is times 0.001mm. Most of the drift is attributable to a change in temperature. The robot should be allowed to warm up for 30 minutes before use for best repeatability. Even so total drift is less than 0.2mm which is quite respectable for a low cost robot. Given a constant temperature a repeatability of better than +/- 0.05mm could reasonably be expected.