What "Measuring" Robot Accuracy Under ISO 9283 Actually Means

"Is this robot arm accurate?" is a question that demands numbers, not impressions. The international standard that defines those numbers for industrial robots is ISO 9283:1998. Its core insight is distinguishing two metrics that are often conflated — Accuracy and Repeatability.

▶ Real simulation — a Franka Panda picking up a cube and placing it at a target location. The 'Arm: planning → executing → reached' state and position/gripper telemetry are shown. ISO 9283 quantifies the precision of motions like this.

Accuracy ≠ Repeatability

The two are commonly confused, but they are fundamentally different.

Rendering diagram…

• AP (Accuracy of Pose) = difference between the mean arrived position and the commanded position. "How close to the center of the target?"
• RP (Repeatability of Pose) = scatter of repeated arrivals (mean + 3σ). "Does it go to the same place every time?"

An archery analogy: AP is "how close to the bullseye," RP is "how tightly grouped the arrows are." Arrows clustered together but off-center means good RP, poor AP.

Standard Measurement Protocol — Cube Workspace

ISO 9283 specifies the measurement method, not just the metrics. A standard cube (recommended side length) is placed inside the workspace. Five poses (P1 at center + P2–P5 at corners) are measured in 30 cycles each — 150 measurement points per robot.

        P2 ────────── P5
        │               │
        │      P1       │     ← P1 = cube center
        │   (center)    │        P2–P5 = corners
        P3 ────────── P4

Real Numbers from Simulation

Results from measuring both Franka models in imSim using the ISO 9283 protocol:

Robot	Max AP	Verdict (industrial cobot threshold: 10 mm)
Panda	6.08 mm	PASS
FR3	7.29 mm	PASS

Both models achieved a cuMotion path planning success rate of 150/150 (100%) with zero standard violations, clearing the industrial cobot baseline of 10 mm.

💡 An honest caveat — these numbers clear the industrial cobot threshold, but they differ from Franka's catalog specification (repeatability ±0.1 mm). Closing that gap would require dedicated precision work (multi-seed IK, etc.) and is a separate effort.

⚠️ "It moved" ≠ "standard pass" — picking up an object is not a PASS. AP and RP must clear the thresholds, there must be sufficient measurement points (n ≥ 150), and violations must be zero. That strict gate (strict_pass) is covered in a separate post.

In a nutshell

📌 ISO 9283 measures robot accuracy as AP (how close to target center) and RP (how consistently the same spot) using a standard cube, 5 poses, and 30 cycles. imSim measured values: Panda 6.08 mm / FR3 7.29 mm — PASS at the industrial cobot threshold.