Automated learning for parameter optimization of robotic assembly tasks utilizing genetic algorithms

A challenge for automating mechanical assembly is that cumulative uncertainties typically exceed part clearances, which makes conventional position-based tactics unsuccessful. Force-based assembly strategies offer a potential solution, although such methods are still poorly understood and can be difficult to program. In this paper, we describe a force-based robotic assembly approach that uses fixed strategies with tunable parameters. A generic assembly strategy suitable for execution on an industrial robot is selected by the programmer. Parameters are then self-tuned empirically by the robot using a genetic-algorithm learning process that seeks to minimize assembly time subject to contact-force limits. Results are presented for two automotive part assembly examples using ABB robots with commercial force-control software, showing that the approach is highly effective and suitable for industrial use.