Payload, reach, repeatability: what the headline numbers actually mean

A plant manager in Ohio once asked a cobot vendor why a robot that listed ±0.02 mm repeatability was dropping parts outside a ±0.15 mm tolerance window — reliably, several times a shift. The vendor's answer was technically accurate and practically useless: the ±0.02 mm figure was measured at reduced speed, reduced payload, at a specific thermal equilibrium, at five points inside the ISO test cube.

None of those conditions matched production.

If you are comparing datasheets across vendors, the three numbers printed at the top of every spec sheet — payload, reach, and repeatability — are a starting point, not an answer. Each one hides enough caveats to reverse a purchasing decision. Here is what those numbers mean, what conditions produce them, and what questions to ask to get to usable information.

Payload: The number you think you know

Rated payload is the mass the robot arm can carry at its tool center point (TCP) while maintaining its advertised speed and repeatability. A UR10e lists 12 kg. A FANUC CR-7iA lists 7 kg. Those figures look simple; they are not.

The wrist moment problem. Payload ratings assume the load is centered at the TCP. The moment — the rotational force the load exerts on the wrist joint — is just as important as the mass. A 6 kg EOAT (end-of-arm tooling) that extends 150 mm past the TCP can exceed the wrist moment rating of a robot with a 10 kg payload rating. Vendors publish maximum allowable moment in N·m alongside payload; buyers routinely ignore this.

The payload curve. Rated payload applies at the robot's minimum reach — typically directly in front of the base, with minimal arm extension. At full reach, permissible payload drops, sometimes dramatically. A robot rated for 10 kg at 500 mm reach may only tolerate 6 kg at 1,200 mm reach because the joint torques at full extension require the controller to throttle speed and acceleration limits to stay within structural margins. Most vendors publish a payload-reach curve in the back pages of the integration manual. Ask for it before the sales call ends.

EOAT is payload too. Whatever your gripper, camera, force-torque sensor, or other tooling weighs counts against the payload rating before you place a single part. A 2.5 kg vacuum gripper on a robot with a 3 kg payload rating leaves you 500 grams of actual working capacity. Factor in tooling weight at the design stage, not the commissioning stage.

The practical rule: Operate at no more than 80% of rated payload under normal production conditions. This is not a rule of thumb — it is a recommendation that appears in several vendors' integration guides because sustained operation at rated payload maximizes joint wear and reduces service life.

Reach: Maximum envelope versus useful envelope

Reach is the maximum distance from the robot's base to the furthest point the TCP can physically travel. A 1,300 mm reach sounds like a generous work envelope. In practice, the useful portion of that envelope — where the robot can move freely in all directions at rated speed — is considerably smaller.

Singularities eat your workspace. All six-axis robots have singular configurations: joint angles where two or more axes align and the robot loses a degree of freedom. Near a singularity, path planning becomes unpredictable, speed must be reduced, and some controllers throw a fault rather than attempt the move. The most common industrial singularities are the overhead singularity (arm fully extended directly above the base), the wrist singularity (joints 4 and 6 aligned), and the elbow singularity (arm fully extended). Vendor reach specifications say nothing about how close to those singularities your intended motion paths will run.

The actual working envelope is a torus, not a sphere. Six-axis robots cannot reach the zone directly behind the base or directly below it (where the shoulder joint mounting interferes). The reachable volume looks like a donut with a cylindrical hole in the center and dead zones at the extremes. If your part pickup and placement points happen to sit near those dead zones, a robot with a 1,400 mm reach may fail to execute the required path while a robot with a 1,100 mm reach in a different mounting orientation succeeds cleanly.

Match reach to your actual path, not your part dimensions. A common mistake: buyers measure the distance between part pickup and placement and specify a robot whose reach comfortably covers that distance. But robots move in arcs, not straight lines. The TCP may need to swing out past the envelope to execute a collision-free path between two points that are geometrically close together. Simulate the actual motion path — most vendors offer free offline programming tools — before finalizing the reach specification.

Repeatability: The most misunderstood spec in robotics

Repeatability is the ability of a robot to return to the same position from the same direction, under the same conditions, repeatedly. The UR10e spec sheet lists ±0.1 mm. The FANUC CR-7iA lists ±0.03 mm. Those numbers come from a specific, controlled test defined by ISO 9283:1998.

Understanding the test is essential to understanding the number.

How ISO 9283 measures repeatability

ISO 9283 defines a Pose Repeatability (RP) test conducted inside the ISO test cube — the largest cube that fits inside the robot's working envelope. The robot moves to five specific points within that cube, approaching each point 30 times from the same direction, and a laser tracker or external measurement system records where the TCP actually lands each time. The repeatability figure is the radius of the sphere that contains 95% of the recorded positions.

This sounds rigorous. It is, on its own terms. But the test conditions are:

• Full rated payload or reduced payload — ISO 9283 allows testing at the manufacturer's choice. If the spec sheet does not state the payload at which repeatability was measured, ask.
• Specified ambient temperature — typically 20°C ±2°C. Thermal expansion of robot links as the arm warms up from cold start shifts repeatability measurably. A robot that achieves ±0.05 mm after a 20-minute warm-up may show ±0.15 mm in the first five minutes of a morning shift.
• Single approach direction — the test always approaches the target point from the same direction. Repeatability measured with bidirectional approach (coming from opposite sides of the point) is called unidirectional versus bidirectional repeatability and the figures can differ by a factor of 2–5x for some robot designs.
• No process forces — the ISO test has no external loading on the TCP during the measurement. In a deburring, welding, or assembly application, process forces push back against the arm and shift the actual TCP position. Compliance (the arm's deflection under load) is a separate specification rarely printed on datasheets.

Accuracy is not repeatability

These two terms are frequently conflated, including in vendor marketing. They are not the same.

Repeatability measures consistency: can the robot hit the same point over and over? Accuracy measures trueness: does the robot actually go where its joint encoders claim it is going?

Industrial robots are highly repeatable and often surprisingly inaccurate. A robot might return to the same position with ±0.02 mm repeatability but be off by ±0.5 mm in absolute space because of kinematic calibration errors, link flexion, and encoder resolution limits. In most production applications, this does not matter — you teach the robot the exact point by moving it there, so accuracy is irrelevant. In offline programming (where you define positions from a CAD model and expect the robot to execute them without teach-in), accuracy matters enormously and most vendor datasheets leave you completely uninformed.

A real comparison: UR10e vs FANUC CR-7iA

The FANUC CR-7iA's ±0.03 mm repeatability looks dramatically better than the UR10e's ±0.1 mm. Before you conclude that the FANUC is three times more precise, ask: at what payload? At what temperature? If the FANUC's number was measured at 2 kg and the UR10e's at 10 kg, the comparison tells you very little about how they will perform at the same production conditions.

In practice, the CR-7iA's ±0.03 mm precision is meaningful for small-part electronics assembly and tight-tolerance inspection. For most material handling, machine tending, or packaging applications where tolerances are measured in millimeters rather than microns, the UR10e's ±0.1 mm is entirely sufficient and the payload and reach advantages dominate the decision.

The five questions to ask every vendor

Once you understand how these numbers are produced, the useful follow-up questions become obvious:

1. "What payload was used for the repeatability test?" If the answer is "rated payload," that is the correct answer. If they can't tell you, the number isn't useful.

2. "What is the repeatability at my specific payload?" If your EOAT plus part weighs 8 kg and the robot is rated for 10 kg, ask for measured data at 8 kg or a calculation of expected degradation.

3. "What is the payload-reach curve?" The table that shows how permissible payload varies with extension. Every integration manual has it. Ask for it as a condition of evaluation.

4. "What is the wrist moment rating, and does my EOAT design stay within it?" Give them your gripper specs and ask them to confirm. If they cannot do this on the sales call, that tells you something about their applications team.

5. "What repeatability should I expect in the first 30 minutes of a cold shift?" Thermal drift is real. Some vendors have this data; many don't. If they don't, ask for a reference site that runs a cold morning shift with similar tolerances.

What good looks like

A vendor who knows their product will walk you through the payload-reach curve unprompted, specify the test conditions alongside the repeatability figure, and help you size the EOAT so you do not cut it too close on wrist moment. They will also tell you the repeatability specification is irrelevant for your application if your tolerances are loose — and explain which spec actually matters for what you are building.

A vendor who cannot answer "at what payload was repeatability measured?" is selling you a number, not an application.

The headline specs exist to help narrow the field. They are not the final answer. Get to the integration manual, the payload-reach curve, and a reference site before you sign anything.

Next in this series: IP ratings explained: when IP54 isn't enough