Robot Repeatability vs Accuracy vs Resolution: Which Spec Matters for Your Application

Robot datasheets quote a single number for “accuracy” — typically something like ±0.02 mm or ±0.05 mm. Most robot datasheets prominently quote repeatability, while absolute positioning accuracy may be absent or separately specified. These are different metrics that measure different things, and confusing them leads to buying a robot that cannot do the job — or overpaying for a robot that exceeds the requirement.

This article exists because the single “accuracy” number on a robot datasheet is typically repeatability — a different metric from absolute positioning accuracy. Buyers who need absolute accuracy (for offline programming) cannot rely on the datasheet number. This article covers the three metrics defined by ISO 9283, why they matter differently by application, and what to verify before procurement. For robot programming method selection (drag teaching vs offline programming), see Article 13.

The Three Metrics: Defined by ISO 9283

ISO 9283:1998 — “Manipulating Industrial Robots: Performance Criteria and Related Test Methods” — is the international standard that defines how robot positioning performance is measured and reported. The standard defines several performance criteria; the three most relevant for procurement are:

Repeatability

Repeatability measures the robot’s ability to return to the same position and orientation over multiple attempts, when approaching from the same direction. It quantifies random error — the scatter of actual positions around the mean position when the robot is commanded to go to the same point repeatedly.

If you teach a robot to go to a point by hand (drag teaching or pendant teaching), and command it to return there 50 times, the spread of those 50 actual positions is the repeatability. The systematic error (the offset between the taught point and where the robot actually goes on average) cancels out — because you taught the point there, and the robot returns to the taught point.

Absolute Positioning Accuracy

Absolute accuracy measures the difference between a commanded position (a position calculated and sent to the robot, not taught) and the actual achieved position. It quantifies systematic error — the offset between where you told the robot to go and where it actually goes.

If you tell the robot (via offline programming or a calculated coordinate) to go to X=500, Y=200, Z=300, and the robot actually goes to X=497, Y=203, Z=298, the absolute accuracy is roughly ±3 mm in this example. This error does not cancel out over multiple attempts — it is a consistent offset caused by mechanical tolerances, gear backlash, link length deviations, and joint encoder offsets.

Resolution

Resolution is the smallest increment of motion the robot can execute. It is determined by the joint encoder resolution, the gear ratio, and the control system’s minimum controllable motion step. If a joint encoder has 1,000 counts per revolution and the gear ratio is 100:1, the theoretical resolution at the joint is 1/100,000 of a revolution — about 0.00006 degrees, or roughly 0.001 mm at 1 meter reach.

In practice, resolution is limited by mechanical factors (stiction, friction, backlash) that prevent the robot from achieving its theoretical encoder resolution. The effective resolution — the smallest motion the robot can reliably produce — is typically larger than the theoretical encoder resolution.

Terminology Table

Actual values vary by robot model, payload, and workspace position. The buyer must define the required value; the supplier must state the quoted metric and test conditions.

Why These Metrics Matter Differently by Application

Pick-and-Place (Taught Positions)

When a robot is taught positions by hand (pendant or drag teaching), repeatability is the critical metric. The operator teaches the robot where to go; the robot returns to that taught point. Systematic error is irrelevant — it is part of the taught position. The only thing that matters is how consistently the robot returns to the taught point, which is repeatability.

What to verify: The required repeatability depends on the part tolerance and process. Define the required positioning consistency and verify the robot’s ISO 9283 repeatability at the relevant workspace position and payload.

Offline Programming (Calculated Positions)

When a robot is programmed offline — the positions are calculated in a simulation and downloaded to the robot — absolute accuracy becomes critical. The robot must go to positions it was never taught, based on a mathematical model. If the model does not match the physical robot (and it never matches exactly), the robot will miss the target.

What to verify: Absolute accuracy varies by robot model, workspace position, and calibration state. If the datasheet does not quote absolute accuracy separately, request it. If not available, ask whether calibration is offered and what accuracy improvement it provides.

Machining with a Robot

When a robot is used for machining (drilling, milling, deburring, sanding), the robot must follow a continuous path accurately. Both path accuracy (deviation from the commanded continuous path) and absolute accuracy matter. The robot must reach calculated positions along a continuous trajectory, not just discrete points.

What to verify: Define the tolerance the robot must maintain along the continuous path. Verify path accuracy and absolute accuracy with the supplier. Robot machining is viable for applications where the tolerance is achievable given the robot’s absolute accuracy — but the specific tolerance depends on the robot model, calibration state, and process.

Assembly (Peg-in-Hole)

When a robot must insert a peg into a hole (a common assembly task), precision insertion involves multiple factors — not a single metric. Robot repeatability, absolute/system accuracy, TCP/fixture error, compliance (e.g., Remote Center Compliance devices), force/torque sensing, vision guidance, search strategies, and the clearance/tolerance between the peg and hole all contribute to whether insertion succeeds. Controller resolution is one factor but rarely the sole or even primary determinant in procurement decisions. Force/torque sensing and compliance are often more important than resolution for practical peg-in-hole assembly.

What to verify: The robot’s repeatability at the insertion position, the TCP calibration accuracy, whether the application uses force/torque sensing or RCC, and the clearance between peg and hole. Ask the supplier to demonstrate insertion with your actual parts and tolerances.

Application-to-Metric Matrix

Requirements are application-dependent. The buyer must define the required positioning performance for their specific application; the supplier must state the metric being quoted and the test conditions under which it was measured.

The Datasheet Trap: “Accuracy” Usually Means Repeatability

Most robot datasheets prominently quote repeatability, while absolute positioning accuracy may be absent or separately specified.

This is not deceptive — repeatability is the better number, and it is the metric that matters for the most common robot application (taught pick-and-place). But it means that a buyer who needs absolute accuracy (for offline programming) cannot rely on the datasheet number.

How to Verify Which Metric Is Quoted

1. Check the label — if the datasheet says “repeatability” or “repeatability (ISO 9283),” it is repeatability. If it says “absolute accuracy” or “positioning accuracy,” verify whether it refers to ISO 9283’s absolute accuracy definition or a different measurement.
2. Check the test condition — ISO 9283 specifies test conditions (position in workspace, load, velocity, number of cycles, approach direction). If the datasheet does not specify test conditions, the quoted number may be a best-case measurement at an optimal workspace position.
3. Ask the supplier directly — “Is the quoted positioning specification repeatability or absolute accuracy per ISO 9283? Under what test conditions was it measured?”
4. If absolute accuracy is not quoted — it is likely not available as a standard specification. Many robot manufacturers do not quote absolute accuracy because it varies with calibration state and workspace position. You may need to request a calibration quote separately.

Calibration: What It Can and Cannot Do

Calibration improves absolute accuracy by measuring the actual robot kinematics and correcting the controller’s model. It does not improve repeatability.

What Calibration Does

• Measures the actual link lengths, joint offsets, and gear characteristics of the specific robot
• Creates a corrected kinematic model in the controller
• Reduces the systematic error between commanded and actual positions — the improvement depends on the calibration method, equipment, and robot model

What Calibration Cannot Do

• It cannot reduce random error (repeatability) — random scatter is caused by mechanical clearances, friction variation, and control system noise
• It cannot compensate for changes that occur after calibration (thermal expansion, wear, mechanical shifts from collisions)
• It cannot improve resolution — that is a hardware property of the encoders and gears

Calibration Methods

Leica LTD 500 is one example; equivalent laser tracker systems from other manufacturers are available. | Kinematic parameter identification | Mathematical model fitting | Depends on model and fitting quality | Hours to days | General OLP improvement | | Tool Center Point (TCP) calibration | Reference point (pointer, sphere) | Improves TCP accuracy | Minutes | Correcting tool offset errors | | Robot-robot calibration (dual robots) | One robot measures the other | Varies | Hours | Multi-robot cells |

Improvement depends on robot model, calibration equipment, and operator expertise. Request the supplier’s stated improvement for your specific robot model.

TCP (Tool Center Point): The Hidden Compounding Factor

All three metrics — repeatability, absolute accuracy, and resolution — are affected by the TCP definition. The TCP is the point at the end of the end-effector where the robot’s positioning is measured and controlled. If the TCP is defined incorrectly (the tool offset is wrong), every position the robot goes to will be offset from the intended position.

TCP errors compound with the robot’s own positioning error. A robot with ±0.05 mm repeatability and a 2 mm TCP offset error will place parts at ±0.05 mm repeatability — but 2 mm away from where they should be.

Practical note: TCP calibration should be performed after every tool change, after any collision, and periodically during operation. The frequency depends on the application tolerance and the risk of tool offset changes (vibration, thermal expansion, mechanical loosening).

Spec-Sheet Worksheet: What to Extract from a Robot Datasheet

Illustrative Scenario: The Offline Programming Surprise

An integrator built a robotic drilling cell for an aerospace component. The robot was specified with “positional accuracy ±0.03 mm” — a number that looked excellent for the ±0.5 mm drilling tolerance requirement. The cell was programmed offline using a CAD model of the part and the robot.

During commissioning, the first drilled holes were 1.2–1.8 mm off from the programmed positions. The ±0.03 mm specification was repeatability — the robot could return to a taught point within ±0.03 mm. But the offline program commanded calculated positions, and the robot’s absolute accuracy was ±1.5 mm without calibration. The systematic error between the simulation model and the physical robot caused the offset.

A laser tracker calibration was performed, measuring the robot’s actual kinematic parameters and updating the controller model. After calibration, absolute accuracy improved to ±0.2 mm — within the drilling tolerance. The cell required periodic recalibration (every 3 months) to maintain accuracy as thermal conditions and mechanical wear shifted the kinematics.

This is an illustrative scenario based on common robot integration commissioning issues. Actual accuracy values depend on robot model, calibration method, and environmental conditions.

Academic Context

Academic research on ISO 9283 provides additional context:

• ISO 9283 repeatability can be computed across the entire workspace using stochastic ellipsoid theory, providing a map of repeatability variation (Springer)
• Robot performance evaluation systems use laser trackers (e.g., Leica LTD 500) to measure ISO 9283 criteria with high precision (Springer)
• Parallel robots generally achieve better repeatability than serial robots when the design is optimized — but this is a design-dependent comparison, not a universal rule (Springer)
• Hexapod robots’ positional capability has been assessed using a procedure adapted from ISO 9283 (Springer)

These findings reinforce that positioning performance varies across the workspace and depends on robot architecture — a single datasheet number does not characterize the robot’s performance everywhere.

Related Articles

This article covers robot positioning performance metrics. For robot programming method selection (drag teaching vs offline programming), which determines whether repeatability or absolute accuracy is the critical metric, see Article 13.