Industrial robot absolute positioning accuracy degrades from sub-0.1 mm to 2-8 mm over weeks of operation due to thermal drift, gear wear, and joint encoder backlash, requiring costly recalibration downtime
technologytechnology0 views
Industrial robot arms (FANUC, ABB, KUKA, Yaskawa) have excellent repeatability (0.02-0.05 mm) but much worse absolute positioning accuracy that degrades from a calibrated baseline of <0.1 mm to 2-8 mm over weeks or months of continuous operation. The drift is caused by thermal expansion of links and gearboxes during operation (up to 50 degrees C temperature rise in harmonic drives), progressive backlash growth in reduction gears, and elastic deformation under varying payloads. For applications requiring absolute accuracy -- offline-programmed welding paths, multi-robot coordination, aerospace drilling -- this drift forces periodic recalibration using expensive laser tracker systems at $2,000-$10,000 per session plus hours of production downtime.
Why it matters: modern manufacturing demands offline programming (generating robot paths from CAD models without manual teaching) to reduce setup time for high-mix production, so offline-programmed paths depend on the robot's absolute accuracy matching the CAD coordinate frame, so when accuracy drifts by 2+ mm the programmed paths produce defective welds, misaligned holes, or collision-triggering position errors, so manufacturers must either frequently recalibrate (expensive downtime) or fall back to manual teach-pendant programming (defeating the purpose of offline programming), so high-mix manufacturers who most need automation flexibility are precisely the ones most burdened by calibration drift.
The structural root cause is that industrial robots use strain-wave (harmonic drive) and cycloidal reducers that inherently exhibit compliance, backlash, and friction hysteresis that change with temperature and wear, and while the joint encoders measure motor-side position accurately, they cannot observe the actual link-side position after the reduction stage, so the controller operates on an inaccurate kinematic model that diverges from reality as the mechanical system changes state.
Evidence
ATT Metrology's ARTEMIS calibration platform demonstrated that a typical industrial robot's volumetric error starts at +/-2 mm and can reach 8 mm for larger machines, reducible to <0.1 mm and 0.13 mm respectively after laser-tracker-based calibration. A December 2024 Robot Automation UK analysis confirmed that 'intrinsic errors come from robot components such as part variations, non-linearity from dynamic influence of hysteresis and friction, resolution of joint position sensors, and overtime wear and tear of parts such as motors and gears.' RoboDK's calibration documentation states accuracies of 0.050 mm for small robots and 0.150 mm for medium robots post-calibration, implying significantly worse pre-calibration baselines. A 2025 IEEE JAS comprehensive survey on data-driven robot calibration noted the rise of machine-learning approaches as an alternative to geometric model calibration, acknowledging that traditional methods require expert knowledge and expensive metrology equipment. Sources: attinc.com/news/artemis-robot-accuracy-solution, robotauto.co.uk/2024/12/20/repeatability-and-accuracy, ieee-jas.net/en/article/doi/10.1109/JAS.2025.125237, robodk.com/doc/en/Robot-Calibration-LaserTracker.html.