Kinematic Calibration and Positioning-Error Compensation of Industrial and SCARA Robots

Absolute positioning accuracy decides whether an industrial robot can be programmed offline and trusted in assembly, machining, inspection, and wafer handling, yet most robots are repeatable to a fraction of what they are accurate. This review examines 100 studies, from the 1980s foundations of robot calibration to articles published in 2026, with attention to serial, SCARA, parallel, and hybrid architectures. The literature is organised into six themes: error sources and kinematic error models (modified Denavit–Hartenberg, complete and parametrically continuous, product-of-exponentials, and arm-signature formulations); measurement systems and calibration procedures (laser trackers, interferometers, vision, double ball bars, constraint-based and self-calibration); parameter identification and measurement-configuration optimisation (observability indices, least-squares and Kalman-type estimators); non-geometric errors, namely joint and link compliance and stiffness modelling; calibration of SCARA, parallel, and over-constrained mechanisms; and data-driven compensation with neural networks, deep belief networks, and hybrid model-plus-learning schemes. A final theme links accuracy to dynamics and control. The review finds a clear shift from purely geometric models toward combined geometric, compliance, and learned corrections, and from laboratory metrology toward robot-mounted and vision-based, in-situ calibration. Key gaps include SCARA-specific non-geometric models, limited cross-robot benchmarks, thermal and wear drift, reduced-dataset learning with guaranteed accuracy, and the lack of standard reporting protocols. The paper closes with recommendations for load-aware, self-updating, and verifiable calibration.