ガウス過程回帰に基づく適応型オンライン制御器チューニング

抄録

This paper presents a new online controller parameter tuning method based on Gaussian process regression (GPR) with experimental data collected during the control process. The proposed method adaptively adjusts parameters in response to modeling errors and unknown disturbances, and provides robust tuning against learning uncertainties by utilizing the posterior variance obtained from the GPR. In particular, our method deals with a simplified approximate model for control targets that are difficult to obtain a linear model, such as multi-rotor Unmanned Aerial Vehicles (UAVs), and tunes parameters to approach an ideal controller without elucidating the specific characteristics of the target system. To confirm the effectiveness of the proposed method, we conduct a real-machine verification of PID tuning using a quadrotor UAV.