A Novel Machine Learning System for Industrial Robot Arm Calibration

Abstract

The application of industrial robot arms in intelligent manufacturing is highly vigorous. Generally, robot arms have high repetitive positioning accuracy. However, they frequently suffer from large absolute positioning error, which can not be directly adopted in high-precision production activities, like chip and cell phone manufacturing. To address this critical issue, we first propose a novel cubic interpolated beetle antennae search (CIBAS)-based robot arm calibration algorithm. The main ideas are three-fold: a) developing a novel CIBAS algorithm to address the local optimum and unstable searching process encountered by the beetle antennae search; b) adopting a particle filter (PF) to suppress the noises in robot arm calibration; c) proposing an efficient CIBAS-based calibration method to search the optimal kinematic parameters. Empirical studies on an HSR JR680 robot arm demonstrate that compared with advancing calibration algorithms, the maximum error of the proposed PF-CIBAS is 21.43% lower than that of the most accurate CIBAS algorithm. Hence, the proposed algorithm is appropriate for a robot arm.