Evolutionary algorithms in robot calibration

ABSTRACT

Robots are universal mechanical systems that are now ubiquitous in manufacturing. One of the most important properties of industrial robots is their kinematic accuracy. Robot’s accuracy is influenced by many factors including manufacture accuracy of mechanical parts and other aspects. Calibration is a technique that allows to identify design and other parameters of the robot to achieve its highest accuracy. There are widely used traditional kinematic calibration methods based on kinematic models of the robot. Simulation is used to compare results of traditional calibration method and a newly developed method based on multi-objective deep learning evolutionary algorithm. EvoDN2 was used together with a reference vector-based evolutionary algorithm, cRVEA, used for optimization, in order to find optimal estimates of the robot parameters. It is well known that the evalutionary algorithms are capable of dealing with noisy data from measurement. Results and comparison of both techniques are discussed and evaluated.

KEYWORDS:

• Robots
• calibration
• evolutionary computation
• genetic algorithms
• data-driven modeling
• multi-objective optimization
• EvoNN
• EvoDN2
• BioGP

Disclosure statement

No potential conflict of interest was reported by the author(s).