When Your Robot Doesn't Go Where You Tell It To

The symptom

You program a position, and the robot goes close — but not exactly there. Maybe it's off by half a millimeter, maybe more. You see it in parts that don't align, in offsets you have to manually correct, in teach points that seem to drift between shifts. The problem gets worse when you switch tools, change payloads, or when the temperature in the cell changes over the day.

Some days are better than others. The first shift after a weekend might be different from midweek. And if you try to use offline-programmed positions instead of manually teaching every point, the errors are noticeably worse.

What's happening

Every robot has two accuracy specs that most people don't distinguish. Repeatability is how consistently it returns to a point you taught it — that's the ±0.03 mm number in the brochure. Accuracy is how close it gets to a position you told it to go to mathematically — and that's often ±0.5–1.5 mm or worse.

The difference exists because the robot's internal model of itself doesn't perfectly match the physical arm. Manufacturing tolerances, wear, temperature changes, and payload effects all contribute. The robot faithfully executes what its controller thinks is the right motion — but its controller's model is slightly wrong.

How to fix it

Software-based accuracy improvement works by correcting the robot's internal model to match the actual arm. A calibration procedure measures how the arm really moves, identifies the errors, and generates a corrected model. From that point on, every commanded position is more accurate.

For errors that change in real time — thermal drift, payload effects, joint compliance — dynamic compensation runs continuously, adjusting commands on the fly to account for conditions that static calibration can't capture.

Both approaches work through the robot's existing command interface. No new hardware, no controller replacement, no mechanical modifications.

Results

• Positional errors reduced from millimeters to sub-millimeter in calibration studies

• Up to 5.7x TCP accuracy improvement on collaborative robot platforms

• Correction persists across production runs — no per-shift recalibration needed for static errors

• Dynamic compensation handles the day-to-day and within-shift variation that frustrates operators

Example

A production cell using offline programming finds that programmed positions are consistently 0.8 mm off from where they should be, requiring manual touch-up on every new program. After software-based kinematic calibration, the same programs run within 0.2 mm — eliminating the touch-up step and reducing changeover time from hours to minutes.