When I first started picking robot arms for a small assembly cell, I stared at spec sheets for hours wondering why every manufacturer bragged about repeatability numbers but barely mentioned accuracy. The short answer turned out to be the most important lesson of my career: repeatability and accuracy measure two completely different things, and confusing them can wreck a project before you ever bolt the robot to the floor.
This guide explains Repeatability vs Accuracy in Robotics Explained the way I wish someone had explained it to me years ago. You will get clear definitions, the ISO 9283 vocabulary, real numbers from a real example, and the practical fixes I have used to push mediocre cells into production-ready performance. By the end you will be able to read any robot datasheet and know exactly what is being promised, and what is not.
Table of Contents
What Is Robot Repeatability
Repeatability is a robot’s ability to return to the same position, over and over, when given the same command. It is a measure of random error, the spread of hits around an average landing point, not whether that average point is actually the one you asked for. If a robot returns to within plus or minus 0.03 mm of where it last landed 50 times in a row, its repeatability is roughly 0.03 mm.
Under ISO 9283, the standard that governs how we test and report these numbers, repeatability is called pose repeatability (RP). The test sends the robot through a defined cycle, measures the actual end-effector pose at the same commanded point many times, and reports the statistical spread of those measurements. A tighter spread means a more repeatable robot, which is exactly why manufacturers are confident quoting this number on a datasheet.
Repeatability is mostly a property of the mechanical system: encoder resolution, joint stiffness, bearing quality, and how much slack is in the gear train. In our test cell, a small desktop arm showed 0.05 mm pose repeatability, but only after we replaced a worn harmonic drive. Mechanical quality is what holds repeatability together, which is why you can also read our piece on backlash in robot gearing to see one of the biggest mechanical repeatability killers.
What Is Robot Accuracy
Accuracy is a robot’s ability to move to the exact commanded position in space, regardless of how consistently it does so. It is a measure of systematic error, the gap between where you told the robot to go and where it actually ends up. If you command a position at coordinates (500.00, 200.00, 300.00) mm and the robot lands at (498.5, 199.2, 300.4), its accuracy at that point is roughly 1.6 mm off target.
ISO 9283 calls this pose accuracy (AP). The standard test commands the robot to several poses inside its workspace, measures where the end effector actually arrives, and compares each measured pose to the commanded pose. The result tells you how close the robot gets to “truth” across the entire workspace, not just how tight the dispersion is at one location.
Accuracy is affected by everything that can bias the kinematic model: link length tolerances, encoder offsets, joint alignment, deflection under load, and thermal expansion. A perfectly repeatable robot can still be wildly inaccurate if its link lengths are slightly off, which is why a six-axis arm off the shipping pallet is almost never accurate to its full theoretical resolution.
Repeatability vs Accuracy: Side-by-Side Comparison
Putting the two metrics next to each other is the fastest way to make the difference click. The table below summarizes the core distinction between repeatability and accuracy, and the bulleted list underneath highlights the engineering consequences.
| Property | Repeatability | Accuracy |
|---|---|---|
| What it measures | Spread of repeated hits (random error) | Deviation from commanded target (systematic error) |
| ISO 9283 name | Pose repeatability (RP) | Pose accuracy (AP) |
| Typical driver | Mechanical quality, encoder resolution, backlash | Kinematic model errors, calibration, deflection |
| What fixes it | Better hardware, tighter bearings, stiffer links | Calibration against external reference |
| Manufacturer guarantees | Yes, on the datasheet | Rarely without calibration |
| Best analogy | Tight cluster on a target | Cluster centered on the bullseye |
Three engineering consequences fall out of that comparison. First, a robot can be highly repeatable and still miss the target by millimeters on every cycle. Second, a well-calibrated robot can be slightly less repeatable than a cheap one, but hit the target more often. Third, when you read a spec sheet, repeatability tells you about the machine, while accuracy tells you about the calibration job someone has to do on top of it.
A Real-World Example: When Repeatability Is Not Enough
One of the clearest examples I have seen came from a 6-axis arm we used in a lab for pick-and-place demos. The datasheet promised a pose repeatability of plus or minus 0.02 mm, which sounds incredible on paper. We sent the arm to a commanded XY target of (400.00, 200.00) mm and recorded 30 cycles.
The hits clustered tightly, with a measured spread of 0.018 mm, almost exactly matching the spec. The problem was the cluster’s center. The average landing point was (401.7, 198.4) mm, a full 2.0 mm away from the true commanded target. The robot was a sharpshooter with a misaligned scope.
That is repeatability without accuracy in a single number. Without external measurement and calibration, the robot would have happily placed every part 2 mm off for the next million cycles. This is the exact scenario where understanding gear ratios in robotics starts to matter, because small kinematic errors compound across every joint and land precisely where you do not want them.
The ISO 9283 Standard for Measuring Both
ISO 9283 is the international standard that defines how industrial robot performance is tested and reported, and it is the reason spec sheets from different vendors can actually be compared. The standard covers pose accuracy, pose repeatability, multi-directional pose accuracy, distance accuracy, pose stabilization, drift, and overshoot, all under a single set of test conditions.
The key definitions worth memorizing are these. Pose accuracy (AP) is the deviation between a commanded pose and the mean of the attained poses when approaching the command from the same direction. Pose repeatability (RP) is the closeness of agreement between attained poses after repeated commands to the same nominal pose. Both are reported at specific points in the workspace, with a defined payload, speed, and ambient temperature.
Because ISO 9283 testing requires external metrology, it is rarely run by end users. Manufacturers run it on new robots in calibrated labs, then publish the numbers. If a datasheet does not cite ISO 9283, the numbers are not directly comparable to anything else on the market, which is something I check before trusting any spec.
What Affects Repeatability and Accuracy
Several engineering factors shape each metric, and they are not the same factors. Knowing which is which tells you where to spend your budget when performance is not where you need it.
Repeatability is driven primarily by mechanical quality. Encoder resolution sets the floor, because each joint can only know where it is to a finite number of counts. Backlash in the gear train lets the joint wobble between directions, which directly inflates random error. Bearing stiffness, link rigidity, and thermal stability all add small contributions, and a worn drive is one of the fastest ways to destroy a previously good repeatability number.
Accuracy is driven primarily by the kinematic model. Link length tolerances, encoder zero offsets, and joint alignment biases are the biggest sources of systematic error. Payload-induced deflection and thermal expansion of the arm add position-dependent bias that changes with where the robot is in its workspace. Calibration against an external measurement system, like a laser tracker or stereo camera, is the standard fix for all of these.
This is also where the surrounding mechanical stack matters. Quality of end effectors in robotics, for example, affects how well the tool center point is defined, which directly limits how accurate the entire system can ever be.
Common Misconceptions: Precision, Repeatability, and Reproducibility
Three related terms get tangled with repeatability and accuracy all the time, and untangling them prevents a lot of wasted arguments on engineering forums. Precision is a casual word that usually means the same thing as repeatability in robotics, but in metrology it can also imply fine resolution. Repeatability is the strict, statistical term: same robot, same command, same conditions, no long time gap. Reproducibility is different: it is the spread you get when conditions change, including different operators, different temperatures, or different days.
Another common belief is that a robot that is highly repeatable is therefore accurate. It is not. The sharpshooter-with-misaligned-scope example above is the canonical counterexample. Yet another misconception is that calibration is optional. Calibration is the only way to convert good repeatability into good accuracy, and skipping it is the single most common reason production cells miss their target tolerances.
Finally, accuracy and repeatability are location-dependent inside the workspace. A robot may show 0.05 mm repeatability near the base and 0.15 mm repeatability at full reach. Always check the datasheet’s measurement points, not just the headline number.
How to Improve Accuracy and Repeatability in Your Robot Cell
Improving these metrics follows a simple priority order. First, lock in repeatability with mechanical quality. Choose encoders with the resolution your application needs, specify low-backlash gearing, and keep harmonic drives or cycloidal gearboxes in good condition. Replace worn cables and belts that introduce hysteresis. A robot that is not repeatable on day one will not be repeatable after calibration either.
Second, calibrate to recover accuracy. Laser trackers, stereo camera systems, and ball-bar tests are the three most common external reference methods. A typical industrial robot can move from 2 to 5 mm of uncalibrated accuracy down to 0.2 to 0.5 mm of calibrated accuracy inside the working envelope. That is a 10x improvement without touching the hardware.
Third, control the environment. Thermal gradients across the arm are one of the largest unmodeled error sources in real cells. Keep the cell climate stable, allow a warm-up period before precision jobs, and re-check calibration after any mechanical intervention, payload change, or major maintenance event.
Fourth, validate the cell with a final measurement. Use a digital probe, vision system, or gauge block to confirm end-effector position at the actual working points. This step is what tells you whether your repeatability plus calibration actually meets the application’s tolerance budget, and it is the part of the workflow most often skipped in a hurry.
Frequently Asked Questions
What is the difference between accuracy and repeatability?
Accuracy is how close a robot gets to the exact commanded target (systematic error), while repeatability is how tightly it clusters repeated hits at the same command (random error). A robot can be highly repeatable but inaccurate if it consistently lands in the same wrong spot.
Are precision and repeatability the same?
In robotics, precision is usually used informally to mean repeatability. Strictly speaking, precision is a broader metrology term that can also imply fine resolution, so it is better to use the term repeatability when comparing robot datasheets.
What is the difference between accuracy, precision, repeatability, and reproducibility?
Accuracy is deviation from the true target. Precision is a casual synonym for repeatability. Repeatability is the spread of repeated hits under the same conditions. Reproducibility is the spread when conditions change, such as different days, operators, or temperatures.
What is the accuracy of a robot?
Robot accuracy is the deviation between the commanded pose and the mean attained pose, measured according to ISO 9283 as pose accuracy (AP). Typical uncalibrated industrial robots sit in the 1 to 5 mm range; calibrated systems can reach 0.2 to 0.5 mm inside the working envelope.
Why do manufacturers guarantee repeatability but not accuracy?
Repeatability is a property of the mechanical system and can be measured reliably on the shop floor, so it is consistent enough to put on a datasheet. Accuracy depends on the kinematic model, payload, environment, and calibration, all of which change once the robot is installed, so it is rarely guaranteed without a calibration step.
Final Thoughts on Repeatability vs Accuracy in Robotics
If you remember one thing about Repeatability vs Accuracy in Robotics Explained, let it be this: repeatability is what the machine does to itself, and accuracy is what the system does against the world. Spec sheets describe the first. Calibration delivers the second. Pair a good repeatable robot with a proper calibration routine and you will get a cell that actually hits the target, not just a cell that misses it consistently.
Your next step is to pull the datasheet for the robot on your floor right now, check whether the repeatability number cites ISO 9283, and schedule a calibration pass if you have not run one in the last 12 months. The improvement in your first article is usually worth more than the cost of the calibration itself.