Saddle Point Optimization with Approximate Minimization Oracle and Its Application to Robust Berthing Control

We propose an approach to saddle point optimization relying only on oracles that solve minimization problems approximately. analyze its convergence property a strongly convex--concave problem and show linear toward the global min--max point. Based analysis, we develop heuristic adapt learning rate. An implementation of developed using (1+1)-CMA-ES as oracle, namely Adversarial-CMA-ES, is shown outperform several existing approaches test problems. Numerical evaluation confirms tightness theoretical rate bound well efficiency adaptation mechanism. As example real-world problems, suggested method applied automatic berthing control under model uncertainties, showing usefulness in obtaining solutions robust uncertainty.