Evaluating Genetic Algorithm based parameter tuning of a black-box object localisation algorithm for random bin picking

Conference: ISR 2018 - 50th International Symposium on Robotics
06/20/2018 - 06/21/2016 at München, Germany

Proceedings: ISR 2018

Pages: 4Language: englishTyp: PDF

Personal VDE Members are entitled to a 10% discount on this title

Verheyen, Maarten; De Maeyer, Jeroen; Demeester, Eric (KU Leuven, Campus Diepenbeek, Dept. of Mechanical Engineering, ACRO research unit, Belgium)

The factories of the future in an Industry 4.0 context are likely to be confronted with the demand for small batch series of individualised products. This requires fast changeovers and re-tuning of hardware and software. Adapting the parameters of software to tune it to novel products is time-consuming, often suboptimal and not very repeatable if performed by humans. Instead, parameters should ideally be estimated automatically. This paper focuses on the automatic tuning of the parameters of a commercially available, black-box computer vision algorithm for recognising and localising products that are randomly positioned in a bin. For this, a tuning approach using a Genetic Algorithm is compared to two baseline approaches, a grid search and a uniform sampling approach. We found that the Genetic Algorithm converged only slightly faster towards the optimal parameters as compared to the uniform sampling approach.