Performance
Accuracy
Accuracy is how close a robot's actual pose gets to the absolute Cartesian pose it was commanded to reach.
What it is
Accuracy measures the gap between where you told the robot to go and where the tool tip actually ends up, in absolute Cartesian terms. Command the TCP to X500 Y0 Z300 and the arm lands 0.8 mm off in X: that miss is the accuracy error.
It matters most when the commanded pose comes from outside the robot, offline programming, a CAD model, a vision-generated coordinate, rather than from a point the operator physically taught. If the robot never visited the point, there's no taught memory to fall back on, so the controller relies entirely on its internal kinematic model to convert Cartesian coordinates into joint angles.
How it actually works
The controller turns a commanded Cartesian pose into joint angles using a kinematic model: nominal link lengths, joint offsets, and zero positions. Every place that model disagrees with the physical arm, a link machined slightly long, gear backlash, deflection under load, a mastering offset that's a hair off, shows up as absolute positioning error. That's why two arms of the same model can differ: their real geometry differs from the shared nominal model by different amounts.
Accuracy has no single published spec the way repeatability does, because it depends on payload, pose, temperature, and how well the arm has been calibrated. A robot can be extremely consistent (returning to the same wrong spot every time) while still being inaccurate in absolute terms. Calibration routines that measure the real geometry and correct the kinematic model are how integrators close that gap for offline-programmed work.
How it differs
- Repeatability · Repeatability is consistency: how tightly the arm clusters when it returns to a single taught point, again and again. Accuracy is correctness: how close it gets to an absolute commanded pose it may never have visited. The FANUC LR Mate 200iD publishes ±0.01 mm repeatability, that number describes scatter around a taught return, not how close the arm lands on a coordinate from a CAD model. An arm can be tightly repeatable and still absolutely inaccurate, because a consistent return to the wrong spot is still consistent.
- TCP offset · A wrong TCP offset produces an accuracy error even on a perfectly built arm: the flange goes exactly where commanded, but the tool tip is measured from the wrong point. Accuracy is the whole chain from commanded pose to real tool tip; the TCP offset is one term in that chain that you set yourself.
Where you meet it in the field
- FANUC LR Mate 200iD · Its ±0.01 mm published figure is repeatability, not accuracy. Use it as the contrast: tight scatter around a taught point says nothing about how close the arm lands on an offline-programmed coordinate.
- Setting the reference position · Absolute positioning depends on a correct reference and mastering. If the zero position is off, every commanded Cartesian pose inherits that offset as an accuracy error.
Common questions
- My robot has ±0.01 mm repeatability, so it's accurate to 0.01 mm, right?
- No. That figure says the arm returns to a taught point within 0.01 mm every time. It says nothing about how close the arm gets to an absolute coordinate it never taught, like one generated by offline programming. A tightly repeatable arm can still be off by a millimeter or more in absolute terms until it's calibrated.
- Why does offline programming expose accuracy problems that teaching doesn't?
- When you teach a point, the arm records its own joint angles, so it can return to that exact configuration regardless of model error. An offline-programmed point is a Cartesian coordinate the arm has never visited, so the controller has to compute joint angles from its kinematic model. Any mismatch between the model and the real geometry lands as a positioning error.
- Can I improve accuracy without changing repeatability?
- Yes, and that's the point of calibration. Measuring the arm's real geometry and correcting the kinematic model reduces absolute error without touching how consistently the arm returns to taught points. Repeatability is a property of the mechanics; accuracy is largely a property of how well the model matches those mechanics.