Data-Driven Task-Specific Optimization of H∞ Controllers via Bayesian Optimization

Tuning controller parameters is crucial for optimizing performance in dynamical systems. In this paper, we propose a two-step approach that combines a model-based method with a data-driven strategy for controller tuning. First, H∞ synthesis is employed to design a controller that ensures conservative and satisfactory performance across a wide range of operating conditions. Then, safe Bayesian Optimization (safeBO) is used to fine-tune controller parameters based on experimental data to enhance performance for a specific task that exceeds the limitations of the H∞ design. The approach is validated through simulations and experiments on a quadrotor, achieving a reduction of up to 60% in position tracking error, demonstrating its effectiveness in safe, efficient, and automated performance optimization.