Constrained Online Optimization Using Evolutionary Operation: A Case Study about Energy-Optimal Robot Control

Abstract

Optimization of full-scale processes during regular production is a challenge that is often encountered in practice, requiring specialized approaches that only introduce small perturbations so that production does not need to be interrupted. Based on a case study, we discuss the potential of Evolutionary Operation (EVOP) derived methods. The case study relates to a badminton robot that has to perform point-to-point motions during a fixed time interval, based on two operation modes: time-optimal motion, which ensures maximum precision but highest energy consumption, and energy-optimal motion, which decreases the energy consumption, but as a trade-off also lowers the precision. The current standard mode of operation is the energy-optimal mode that is constructed from off-line optimization on simulations. An online EVOP steepest ascent optimization to further reduce the energy consumption by fine-tuning the implemented energy-optimal mode was implemented. The constrained nature of the problem, where energy needs to be minimized subject to a time constraint, was transformed to an unconstrained single-objective optimization using Derringer desirability functions. Two important contributions were made: (i) the online optimization of the energy-optimal motion lowered the energy consumption by 4.7% while keeping the precision constant and (ii) the more stringent time-constraints implemented in desirability functions lead to an operation mode with maximum precision and 51.7% less energy consumption than the current time-optimal motion.