Two different questions: does the robot always land in the same spot? And is that spot actually the one you asked for? A robot can nail the first and quietly fail the second.
Repeatability measures how consistently a robot returns to the same taught point, cycle after cycle — tight clustering of its actual stopping positions around each other, regardless of whether that cluster happens to sit where you originally intended. Accuracy measures how close the robot actually gets to the true, commanded coordinate in space — whether the center of that cluster matches where you really wanted it to go. A robot can have excellent repeatability while having mediocre accuracy: it returns to the exact same wrong spot every single time, off from the true target by a fixed amount caused by calibration error, imperfections in the kinematic model, or thermal drift. This is exactly why industrial robots are specified and sold primarily on repeatability, not accuracy — and why that gap causes real problems the moment a robot has to move somewhere it was never physically taught.
Think of a dartboard. Repeatability asks: how tightly grouped are your throws? Ten darts landing within a 2 cm circle of each other is excellent repeatability, whether that circle sits on the bullseye or in the outer ring. Accuracyasks a completely separate question: is the center of that group actually on the bullseye? A tight group off in the 4-ring is repeatable but inaccurate. A loose scatter of throws that happens to average out near the bullseye is accurate — on average — but not repeatable. Robots behave exactly like this: repeatability comes from mechanical consistency (backlash, stiffness, encoder resolution, servo tuning) repeating the same small errors the same way every cycle, while accuracy comes from how well the robot's internal kinematic model and calibration match the physical machine and its real-world environment.
When an engineer teaches a robot a spot-welding point, a pick location, or a dispensing path — by jogging the arm there and recording the joint values, or by offline programming referenced to the robot's own coordinate frame — the "true" target and the taught target are the same thing. There is no independent, external measurement of where the point "really" is; the point is defined by the robot having been there once. Every time the robot returns to that recorded position afterward, all that matters is whether it lands in the same place it landed during teaching. That's repeatability, full stop, and any absolute-accuracy error the robot has gets baked in identically every cycle and simply cancels out.
That's why the bottom-right quadrant above — tight cluster, offset from true target — describes the overwhelming majority of industrial robots as they ship from the factory, uncalibrated for absolute positioning. It's also perfectly adequate for the overwhelming majority of the work those robots do. Manufacturers lead with repeatability because it's the number that predicts success on taught-point tasks, and it's also the easier number to guarantee: a controlled factory measurement of return-to-point consistency, rather than a claim about matching every arbitrary point in 3D space to some absolute reference frame.
The moment a robot has to move to a coordinate it was never physically taught — one computed on the fly by a vision system, generated from CAD data, or handed over from another machine's coordinate frame — repeatability stops being the relevant spec. There is now a genuine, external "true" coordinate the robot has never visited before, and only absolute accuracy determines whether the robot actually gets there.
That headline number is almost always repeatability, not accuracy — and the two can differ by an order of magnitude or more. A robot rated at ±0.02mm repeatability might have absolute accuracy of only 0.2–1mm, or worse, depending on how well its factory-nominal kinematic model matches the physical arm and how much thermal expansion shifts its links over a shift. Repeatability is cheap and flattering to publish: it's measured by sending the robot back to the same taught point thousands of times and reporting how tightly those returns cluster. Absolute accuracy requires an external metrology reference — a laser tracker or coordinate-measuring system — to check the robot against a truly independent ground truth, and that number is almost never the headline spec because it's a less impressive one. If a task requires the robot to hit a computed or externally-supplied coordinate it has never been taught, the repeatability spec tells you nothing about whether it will succeed — you need the accuracy spec, or a calibration/vision-correction step that compensates for the gap between the two.
Explains the two specs most often confused on an industrial robot datasheet: repeatability, which measures how consistently a robot returns to a taught point, and accuracy, which measures how close it gets to the true, independently-verified coordinate. A robot can have excellent repeatability and mediocre accuracy at the same time, which is exactly the normal case for uncalibrated industrial arms — and exactly why manufacturers headline repeatability, not accuracy.
Repeatability is how tightly a robot clusters around the same programmed position across many repeated cycles. It is driven by mechanical factors — backlash in the gear trains, structural and joint stiffness, servo tuning, and encoder resolution — that repeat the same small errors the same way, cycle after cycle. Because the measurement compares the robot only against its own past behavior, repeatability says nothing about whether the position it keeps returning to is the "correct" one by any outside standard.
Accuracy is how close a robot gets to a true coordinate, independently verified by external metrology such as a laser tracker or coordinate-measuring machine. It depends on how well the robot's internal kinematic model — the mathematical description of link lengths, joint offsets, and geometry used to convert a target coordinate into joint commands — matches the physical arm, plus factors like gravity sag, backlash bias, and thermal expansion of the links during operation. A robot can be extremely repeatable while still being measurably inaccurate, because none of the sources of repeatability error prevent the cluster itself from being centered in the wrong place.
The vast majority of industrial robot tasks — spot welding, pick-and-place, dispensing, palletizing — use positions taught to the robot by demonstration or offline programming referenced to the robot's own coordinate frame. In that workflow, the "true" target and the taught target are identical by definition, so all that matters for task success is that the robot returns to the same spot reliably. Repeatability is also easier and cheaper to guarantee at the factory, since it only requires the robot to be tested against itself, not against an external reference frame.
Accuracy matters the moment a robot must reach a coordinate it was never physically taught: a position computed by a vision system locating a randomly placed part, a coordinate derived from CAD data, or a target handed off from another robot or fixture using a different reference frame. In these cases repeatability tells you nothing useful, because there is now a genuine external target the robot has never visited. Vision-guided or externally-corrected systems typically compensate for this by measuring the actual part position live and adjusting the commanded coordinate, rather than relying on the robot's uncorrected absolute accuracy.
Repeatability is how consistently a robot returns to the same taught position over many cycles — tight clustering around itself. Accuracy is how close the robot gets to the true, independently verified coordinate — whether that cluster is centered where it was actually supposed to be. A robot can have excellent repeatability and poor accuracy at the same time.
No. That number is almost always the repeatability spec, measured by sending the robot back to the same taught point thousands of times. Absolute accuracy — how close the robot gets to an arbitrary true coordinate it has never visited — is typically much worse, often 0.2mm to 1mm or more, and is rarely the headline number because it requires external metrology to verify and is a less flattering figure to publish.
Because most industrial tasks use positions the robot was itself taught by demonstration or offline programming referenced to its own coordinate frame. In that setup, the taught point is the true target by definition, so only repeatability — returning to that same point reliably — determines task success. Accuracy against an external reference simply doesn't come into play for most repetitive manufacturing work.
Absolute accuracy becomes the deciding spec whenever a robot must move to a coordinate it was never physically taught — for example, a position computed by a vision system for a randomly placed part, a coordinate derived from CAD data, or a target shared from another machine's coordinate frame. Since there's now a genuine external target the robot has never visited, repeatability alone can't predict whether it will succeed.
Repeatability comes from mechanical consistency — backlash, joint stiffness, and encoder resolution repeating the same small errors identically each cycle. Accuracy failures come from a different source entirely: calibration error, imperfections in the kinematic model used to convert coordinates into joint angles, gravity sag, or thermal drift as the arm heats up during operation. None of those accuracy-degrading factors interfere with the robot's ability to return to the same (wrong) spot consistently.
Often, yes. Absolute accuracy can typically be improved through calibration routines that measure the real robot against an external reference (such as a laser tracker) and correct the kinematic model to compensate for the deviation, or through real-time vision-guided correction that measures the actual target and adjusts the commanded coordinate on the fly, rather than trusting the robot's uncorrected internal model.
Try our Robotics & Automation Studio
More calculators, simulators, and guides for this discipline.