ISO 9283 Repeatability: How to Evaluate Cobot Positioning Performance

A cobot’s repeatability value is useful only when the measurement conditions are known. A catalog figure such as ±0.02 mm may indicate strong mechanical consistency, but it does not by itself establish how the robot behaves at different reaches, under a production payload, after warm-up, or when approaching a target from changing directions. ISO 9283 provides a controlled framework for separating these questions.

The central distinction is simple but commercially important: repeatability is not accuracy. Repeatability describes how closely a robot returns to the same achieved pose when it is commanded to repeat an identical movement under defined conditions. Accuracy describes how close that achieved pose is to the commanded pose. A robot can therefore be highly repeatable while maintaining a consistent offset from its programmed coordinates. For dispensing, welding, polishing, and machine tending, repeatability may be the principal concern. For offline programming, fixture interchangeability, calibrated inspection, or absolute positioning relative to external equipment, accuracy and calibration stability become equally important.

What ISO 9283 actually evaluates

ISO 9283, Manipulating industrial robots — Performance criteria and related test methods, defines terminology, performance characteristics, and test methods for manipulating industrial robots. It is applicable to cobots as a class of industrial robot, but it is not a cobot-specific safety standard and should not be interpreted as a safety certification.

The standard covers more than point-to-point positioning. Its performance framework includes pose and distance accuracy, pose and distance repeatability, overshoot, path performance, velocity-related behavior, settling characteristics, static compliance, and weaving performance where relevant. Not every characteristic carries equal weight in every application. A palletizing cell may be mainly concerned with reach-dependent repeatability, cycle time, and payload. A robotic inspection or metrology process may require more attention to absolute accuracy, path accuracy, and the relationship between robot coordinates and an external measurement frame.

For cobot comparisons, the ISO 9283 repeatability discussion generally centers on the robot’s ability to reproduce a programmed pose. The measured result is influenced by joint transmission behavior, encoder resolution, structural stiffness, controller interpolation, servo tuning, thermal condition, cable routing, payload inertia, and the selected test path. That is why a single repeatability number should be treated as a result of a specified test, not as a universal property that remains unchanged across all tasks.

Repeatability, accuracy, and the “±” notation

Manufacturers frequently state repeatability in millimeters with a plus/minus sign. This convention is easy to read but can conceal important methodological differences. Before comparing values, confirm whether the declaration refers to ISO 9283 pose repeatability, an internal factory method, a best-case point in the work envelope, or a value derived from a particular payload and cycle condition.

Under ISO 9283-style testing, the robot is repeatedly moved to a target pose. The actual achieved positions are measured over a series of repetitions. The collection of measured points has a center, or mean position, and a scatter around that center. Repeatability is based on this distribution of returned positions rather than the gap between the commanded point and the mean achieved point.

Accuracy asks a different question: where is the center of that point cluster relative to the programmed target? If all return points lie tightly together but are displaced by 0.5 mm from the target, repeatability can be excellent while accuracy is poor. A calibration routine may improve absolute accuracy substantially without changing the underlying repeatability by the same amount.

The practical implication is that a ±0.03 mm declaration is not automatically superior or inferior to ±0.05 mm. The difference may be meaningful, negligible, or not directly comparable, depending on the test configuration and the tolerance stack of the process.

The measurement conditions are part of the result

ISO 9283 is valuable because it treats testing as a system of controlled conditions rather than a single endpoint reading. Positioning performance must be evaluated within a defined test volume and at specified poses. The robot is not assessed only in the mechanically favorable area closest to its base. A meaningful evaluation samples the portion of the work envelope relevant to the intended task.

For articulated cobots, geometry matters. At extended reach, structural deflection and joint compliance generally have more influence than near the base. Near kinematic singularities, a small Cartesian movement can require disproportionate joint motion, and controller behavior may differ from that observed in a central working zone. A repeatability value measured at one point cannot establish equal performance throughout the whole reachable envelope.

Payload conditions also require close attention. The mass of the tool is only one part of the load case. Center of gravity, moment of inertia, cable drag, hose stiffness, and off-axis loading can affect settling and positional behavior. A lightweight gripper mounted close to the flange is not equivalent to a long, offset screwdriver, a vision head, or a dispensing assembly with moving hoses. If the production end effector differs materially from the declared test load, the catalog number should be considered a baseline rather than an acceptance value.

Temperature and duty cycle are another source of confusion. Gearboxes, motors, brakes, and links change condition as the robot warms. A short demonstration involving isolated point moves can produce a different result from a continuous production sequence. Where dimensional control is critical, the evaluation protocol should define warm-up, speed and acceleration settings, dwell or settling time, environmental conditions, payload, and the number of cycles. These details make a result reproducible.

How an ISO 9283 positioning test should be read

A positioning test begins with a defined robot configuration, test payload, and measurement arrangement. The robot moves to selected target poses repeatedly. At each target, a measuring system records the achieved position; orientation may also be evaluated depending on the test objective. The repeated measurements reveal both the average attained position and the spread of returns around it.

The measurement system must be capable of resolving the performance being claimed. If a robot is being assessed for repeatability in the hundredths-of-a-millimeter range, a coarse indicator or an inadequately controlled vision setup cannot provide a defensible result. Laser trackers, high-accuracy optical systems, calibrated dial-based arrangements in limited cases, or other traceable metrology solutions may be used depending on the required uncertainty and measurement volume. The critical issue is not the brand of instrument but whether its uncertainty, setup, and coordinate transformation are appropriate for the expected robot performance.

Direction of approach deserves explicit treatment. Backlash, friction, gravity loading, and control compensation can make a robot reach a nominal point differently when approaching from opposite directions. A procedure that always approaches the target through the same trajectory measures one form of repeatability: repeatability under controlled, repeatable approach conditions. That can be entirely valid for a process with a fixed programmed approach. It does not prove equivalent behavior when the application permits varied approach paths or frequent task changes.

Settling time matters for the same reason. Taking the measurement immediately after deceleration can include transient vibration or residual servo motion. Measuring only after an extended dwell may obscure a cycle-time limitation. The correct dwell time is therefore not merely a metrology choice; it must reflect the process requirement. A vision-guided pick station may tolerate a pause. A high-throughput dispensing or inspection path may not.

Why ISO 9283 repeatability is not the same as cell capability

Even a properly reported ISO 9283 cobot repeatability value describes the unloaded or specified-load robot performance under a controlled test method. Production capability depends on the full tolerance chain:

• robot positioning behavior at the required poses and orientations;
• flange, tool, and fixture compliance;
• tool-center-point definition and calibration quality;
• part location variation and fixture repeatability;
• gripper variation, part deformation, and contact force;
• vision calibration, conveyor tracking, or external-axis synchronization where used;
• thermal movement of the cell, process equipment, and workpiece.

Consider a screwdriving application with a hole-position tolerance of ±0.20 mm. A robot repeatability specification of ±0.03 mm looks comfortably smaller, but it does not guarantee successful engagement. The driver can deflect, the bit can float, the fixture can shift, the part may vary, and the target location may be derived from a camera with its own uncertainty. Conversely, a robot with a less impressive catalog repeatability value may perform reliably if the process includes a compliant mechanism, a locating feature, or force-guided search.

For this reason, the appropriate engineering question is not “Which cobot has the smallest repeatability figure?” It is “What portion of the process tolerance is allocated to robot positioning after all other error sources are accounted for?” That allocation should include a margin for operational variation rather than relying on a nominal best-case number.

Common interpretation errors during technical comparison

Using repeatability as proof of absolute placement. A cobot can repeatedly return to a location that is offset from CAD coordinates or from a second cell. Applications requiring coordinate transfer, fixture replacement, or offline-generated paths need evidence of accuracy, calibration method, and recalibration stability.

Comparing values without the tested payload and reach. A published result measured near the center of the workspace with a compact nominal load cannot be assumed at maximum reach with a high-inertia tool. Ask for the declared conditions and for any performance curves or test records available for the intended operating region.

Ignoring orientation. A robot’s tool point can be positioned acceptably while its orientation error affects insertion, sealing, scanning angle, welding torch geometry, or camera focus. Position and orientation must be considered together where the process is sensitive to both.

Equating a point test with path performance. Point-to-point repeatability does not establish that a cobot will maintain a uniform bead, follow a contour accurately, or scan at constant stand-off. Continuous-path work requires evaluation of trajectory behavior, velocity fluctuation, cornering deviation, and dynamic deflection.

Assuming compliance is irrelevant because the robot is collaborative. Collaborative design and force-limited operation do not eliminate structural deflection. In fact, contact processes can make stiffness and force-control behavior more consequential than no-load positioning repeatability.

What to request in an engineering evaluation

A supplier declaration that references ISO 9283 should identify the standard edition and clearly state the specific characteristic being reported. “ISO 9283 compliant” is too broad if no test item, operating condition, or result is provided. The useful evidence is a test report or performance declaration that connects the stated repeatability figure to a defined method.

The technical record should establish the robot model, controller and software version where relevant, test payload and center of gravity, selected test poses, reach region, programmed speed and acceleration, approach conditions, warm-up state, dwell time, number of repetitions, measurement equipment, and reported result format. If the proposed application operates in a special configuration—such as ceiling mounting, a mobile base, a seventh axis, or a long-reach tool—those conditions should be represented in acceptance testing rather than inferred from a standard floor-mounted test.

For a production decision, factory acceptance data and an application-level validation serve different purposes. ISO 9283 provides a disciplined basis for characterizing robot performance. The final cell test should validate the actual critical quality output: insertion success, bead placement, inspection registration, placement error, or another process-specific measure. The first proves what the robot can reproduce under defined conditions; the second proves whether the integrated process meets its functional requirement.

A defensible way to use the specification

ISO 9283 repeatability is most valuable when it is used to narrow uncertainty, not when it is treated as a standalone purchasing shortcut. It gives technical evaluators a common language for asking how a value was obtained, what it represents, and whether two vendor declarations are genuinely comparable.

A low repeatability number supports confidence only when the test volume, payload, approach path, metrology method, and dynamic conditions resemble the intended use. Where process tolerances are tight, the decision should be based on the complete error budget and an acceptance test built around the actual end effector, part presentation, and production cycle. That approach turns a catalog parameter into an engineering decision: not simply whether the cobot can return to a point, but whether it can maintain the required process outcome over repeated real cycles.