Robot repeatability vs accuracy: typical values from ±0.01 to ±0.2 mm
Robot repeatability ranges from ±0.01 to ±0.2 millimetres depending on the design. As a performance figure defined by ISO 9283, it describes how precisely a robot returns to a position it has reached before. It has to be distinguished from absolute accuracy, meaning how well the robot hits a newly programmed position: uncalibrated, that figure is several times worse. This article explains both figures and the measuring method of the standard. It also covers typical values by robot type and the factors that erode accuracy in operation.
Reading time approx. 10 minutes
What robot repeatability means
Repeatability is the scatter with which a robot reaches the same pose again and again. A pose covers the position and orientation of the tool. In the standard test, the robot approaches a taught point 30 times, and repeatability states the radius around the mean value within which these approaches land. A value of ±0.05 millimetres therefore means: the robot hits the same point reliably within a circle of that radius.
For classic teach-in programming, repeatability is therefore the key figure. When you teach positions by hand, you bypass all systematic errors of the robot. The taught point already is the real point, and the robot only has to find it again. Series production with fixed positions, palletizing, machine tending and pick-and-place all work on this principle.
In the datasheet, repeatability appears as a ± value in millimetres, usually measured as pose repeatability to ISO 9283. It is the accuracy figure you find in almost every datasheet. Absolute accuracy is usually missing.
Repeatability vs accuracy: the difference from positioning accuracy and absolute accuracy
Repeatability measures how well a robot finds a known point again. Positioning accuracy measures the deviation between the programmed target position and the point reached. As a rule of thumb, a standard industrial robot has a repeatability of around 0.1 millimetres, but an absolute accuracy of only about 1 millimetre. The term positioning accuracy is often used synonymously with absolute accuracy. The ISO figure pose accuracy also covers orientation.
A robot can be excellent at repeatability and still be systematically off target. Uncalibrated robots are frequently accurate to only 1 to 3 millimetres in absolute terms. In its product sheet on absolute calibration, ABB even puts the difference between the ideal computational model and the real robot at typically 8 to 15 millimetres.
The cause lies in the difference between model and machine. The controller calculates with an ideal kinematic model, while the real arm deviates from it. Manufacturing tolerances in the links, adjustment errors of the axes, gear backlash and deflection under load add up. In teaching this has no consequences, because the real point is stored. Absolute accuracy becomes relevant as soon as positions come from CAD data (CAD stands for computer-aided design) or from offline programming. Those coordinates refer to the ideal model.
Path accuracy: when the path matters, not the point
Path accuracy describes the precision with which a robot follows a programmed path while it is moving. Point accuracy at standstill helps little once the tool works on the move. In welding, gluing, dosing, painting and laser processing, the result forms along the entire path. ISO 9283 therefore distinguishes path accuracy from path repeatability, meaning the scatter of several runs along the path.
The requirements on the path are often stricter than on the point. Precise laser welding demands path deviations below 0.1 millimetres, as Michael Garstenauer of the controller manufacturer Keba explains in the trade magazine all-electronics. Manufacturer data on path accuracy is rare, though. Most datasheets state only pose repeatability, and path values depend heavily on speed and load.
If you are looking for a robot with high path accuracy, comparing datasheets alone therefore gives only a rough indication. What counts are path measurements under your own process conditions or manufacturer data specifically for path-guided processes.
How ISO 9283 measures accuracy
The ISO 9283 standard defines how the accuracy figures of industrial robots are measured so that they are comparable. The test runs inside an imaginary cube in the working space, whose position the standard specifies and whose size depends on the working space. For the pose figures, the robot approaches five measuring points in the test cube 30 times. For the path figures, it runs the test path 10 times, at 100, 50 and 10% of rated speed. Both tests run with a defined load.
From the measurement series, the four central figures emerge: pose accuracy, pose repeatability, path accuracy and path repeatability. Two robots are comparable only at the same cube size, test load, speed and environment.
In practice, note: the datasheet value is produced under test conditions, meaning in a warmed-up state, at a defined load and on the standard path. In real operation, with a cold robot, changing loads and different arm positions, the value is a guide. Only your own measurement in the process delivers a reliable tolerance.
Repeatability states how precisely a robot hits the same point again. Absolute accuracy states how precisely it hits a newly programmed position. Datasheets quote ±0.01 to ±0.2 millimetres of repeatability, while uncalibrated robots are often 1 to 3 millimetres off in absolute terms.
Typical repeatability values by robot type
The range runs from ±0.01 millimetres for SCARA robots and small six-axis robots to ±0.2 millimetres for palletizing robots. Repeatability depends more on size and design than on the manufacturer. Small, stiff robots are more accurate than large long-reach arms with high payload, and cobots (collaborative robots) sit in between. The table shows datasheet values of common models, from the SCARA robot (SCARA stands for Selective Compliance Assembly Robot Arm) to the palletizer:
| Robot type | Typical range | Examples according to datasheet |
|---|---|---|
| SCARA | ±0.01 to ±0.02 mm | Epson GX8: ±0.015 mm (axis 3: ±0.01 mm), Epson T3-B: ±0.02 mm |
| Small six-axis robot (up to approx. 10 kg) | ±0.01 to ±0.03 mm | ABB IRB 1100: 0.01 mm, FANUC LR Mate 200iC: ±0.02 mm (variant 5L: ±0.03 mm) |
| Delta robot | ±0.02 to ±0.03 mm | FANUC M-1iA: from 0.02 mm |
| Cobot | ±0.02 to ±0.05 mm | AUBO i5: ±0.02 mm, FANUC CRX-10iA: 0.04 mm, UR10e: ±0.05 mm |
| Medium to large six-axis robot | ±0.05 to ±0.1 mm | KUKA KR 60 HA (high-accuracy variant): ±0.05 mm, ABB IRB 6400R: 0.1 mm |
| Palletizing robot | up to ±0.2 mm | ABB IRB 460: 0.2 mm |
Some of the cobot values come directly from the Unchained Robotics marketplace, where they appear in the specifications on the product pages, for example for the Agile Robots Yu 5 at ±0.05 millimetres. If you want to compare models, filter the cobot range or the industrial robots directly by their specifications.
The SCARA design stands out: four axes and the stiff vertical guidance deliver values of ±0.01 to ±0.02 millimetres. More on this in the post on how a SCARA robot works with its four axes. The palletizer at the other end of the table also shows that high accuracy is not an end in itself. The fact that ABB specifies its palletizer IRB 460 at 0.2 millimetres ex works reflects the real need of this application.
Are cobots less accurate than industrial robots?
On the datasheet value alone, cobots are on par. A UR10e at ±0.05 millimetres to ISO 9283 is in line with many conventional six-axis robots of medium size. Compact cobots like the AUBO i5 reach ±0.02 millimetres. The widespread claim that industrial robots are fundamentally more accurate is therefore only partly confirmed by pose repeatability.
The real difference lies elsewhere. Industrial robots are built with more stiffness and hold their accuracy even at high speed, full payload and on long paths. Cobots are designed for compliance and safety, and their safety functions limit speed and dynamics. Under process load and in path accuracy this becomes noticeable, even if the catalogue value looks identical. An assessment of both designs with costs and fields of use is given in the overview of cobot advantages, disadvantages and costs.
What affects accuracy in operation
Temperature, payload and speed, gear backlash, arm position and mounting pull accuracy in operation below the datasheet value, which is produced under controlled test conditions:
- Temperature: A cold robot is less accurate. A study in the journal Applied Sciences shows that temperature changes can worsen repeatability by up to one order of magnitude. In a warmed-up state, by contrast, the robot can perform clearly better than its datasheet value. Precision applications therefore schedule warm-up cycles.
- Payload and speed: As load rises, the arm deflects more; as speed rises, dynamic deviations grow. Measurements show deterioration already above around 50% of rated speed.
- Gear backlash: Play in the axis gears acts as a lever across the arm length. Just 3 arcminutes of backlash on an axis with a 500 millimetre lever arm produce around 0.44 millimetres of deviation at the tool.
- Arm position: At the edge of the working space, with the arm fully extended, stiffness drops and with it accuracy. The same task near the base runs more precisely.
- Mounting: A robot on a vibrating frame or an uneven foundation loses accuracy through the substructure alone.
When troubleshooting in operation, you therefore check temperature drift, load changes, mounting and the measurement of the tool centre point first, before the mechanics come under suspicion.
Calibration: from the tool centre point to absolute calibration
Calibration is the alignment between the computational model of the controller and the real machine, and it starts at the tool. The tool centre point (TCP) defines the effective point to which every accuracy figure refers. If the TCP is measured incorrectly, this error affects every position approached. How TCP measurement works step by step is shown in the post on understanding the tool centre point (TCP).
Absolute calibration goes one level higher. A laser tracker measures the real robot in many poses, and the deviations produce a corrected model in the controller. In its absolute calibration, ABB compensates around 40 parameters per robot, from geometry to deflection. According to its own data, ABB achieves an absolute accuracy of around 0.5 millimetres across the entire working range this way. An independent study confirms the order of magnitude on an ABB IRB 1600. Laser tracker calibration there reduced the mean position error from 0.97 to 0.36 millimetres and the maximum error from 2.16 to 0.70 millimetres.
This is necessary only when programmed coordinates have to be correct: in offline programming from CAD data, when swapping robots while keeping existing programs, or in metrology-guided processes. The measuring equipment remains a specialist matter. A laser tracker alone costs over 100,000 euros, so absolute calibration is usually bought in as a service or a factory option. With the HA series, KUKA also builds robots that are adjusted ex works for high accuracy.
How much accuracy your application really needs
Palletizing manages with ±0.2 millimetres, insertion assembly works with component tolerances around 0.5 millimetres, laser welding demands path deviation below 0.1 millimetres. Accuracy costs money, so before choosing it pays to look at the real need of your process:
| Application | Accuracy requirement | Assessment |
|---|---|---|
| Palletizing and packaging | ±0.2 mm is enough | ABB specifies the palletizer IRB 460 at 0.2 mm ex works |
| Machine tending, pick-and-place | uncritical with taught positions | clamping fixtures absorb deviations |
| Joining and assembly | comfortable from ±0.03 mm | insertion assembly works with tolerances around 0.5 mm |
| Welding, gluing, laser processing | path accuracy counts, laser welding below 0.1 mm | the result forms along the path, the pose value is only a guide |
| Measuring, inspection, machining | hundredths of a millimetre to under ten micrometres | requires calibration, external metrology or a different machine class |
In joining and assembly, the right tool plays a part too. An overview is given in the post on what types of end effectors are available for industrial robots.
The order of selection is clear from this. First quantify the process and its tolerance, then choose the suitable design, and compare models by datasheet last. If you do it the other way round and buy the most accurate model first, you often pay for precision that your process never uses.
Frequently asked questions
What does repeatability mean for a robot?
Repeatability states how precisely a robot reaches a position again that it has approached before. It is measured to ISO 9283 and appears in the datasheet as a ± value in millimetres. Typical values lie between ±0.01 millimetres for small robots and ±0.2 millimetres for palletizing robots.
What is the difference between repeatability and absolute accuracy?
Repeatability measures how well a robot finds a taught point again, absolute accuracy how well it hits a newly programmed coordinate. A standard robot has a repeatability of around 0.1 millimetres, but uncalibrated an absolute accuracy of only 1 to 3 millimetres. The reason: the computational model of the controller deviates from the real arm.
How accurate is an industrial robot?
Small six-axis robots and SCARA robots repeat positions to ±0.01 to ±0.03 millimetres. Medium and large six-axis robots reach ±0.05 to ±0.1 millimetres, palletizing robots around ±0.2 millimetres. Uncalibrated absolute accuracy lies clearly above that, often at 1 to 3 millimetres.
Are cobots less accurate than conventional industrial robots?
On the datasheet value, cobots are on par: they reach a repeatability of ±0.02 to ±0.05 millimetres, in line with many six-axis robots. Industrial robots keep their accuracy, however, even at high speed, full load and on long paths, because they are built with more stiffness.
What is ISO 9283?
ISO 9283 is the standard for the performance criteria of industrial robots. It defines, among other things, pose accuracy, pose repeatability and path accuracy. It also specifies the measuring method: five points in a test cube, 30 measuring cycles, defined load and speed. Only under identical test conditions are the values of two robots comparable.
What does absolute calibration achieve?
Absolute calibration measures the real robot with a laser tracker and corrects the model in the controller. After calibration, ABB states an absolute accuracy of around 0.5 millimetres across the entire working range. In an independent study on an IRB 1600, laser tracker calibration reduced the mean error from 0.97 to 0.36 millimetres. It makes sense for offline programming and CAD-based coordinates.
Why does my robot not reach its datasheet accuracy?
The datasheet value applies under test conditions to ISO 9283. In operation, a cold state, temperature changes, high speed, full payload, a fully extended arm position, gear backlash and a compliant mounting all worsen the result. If deviations occur, check temperature drift, load changes, mounting and the measurement of the tool centre point first.