Online Updating Data-Driven Model Predictive Control for Quadrotors in Close Formation Based on Gaussian Process Regression

The formation of quadrotors displays significant potential and has wide-ranging applications, making it an essential research area for autonomous machine operation and multi-agent cooperation. However, as a situation in formation control, controlling the close formation of quadrotors faces the challenge of intricate aerodynamic disturbances. The aerodynamic effects introduce significant formation errors that jeopardize formation safety and are difficult to model due to their high nonlinearity. To address this issue, our paper presents an aerodynamic modeling method that utilizes Gaussian process regression and an online updating data-driven model predictive control algorithm for quadrotors in close formation. We self-develop a large-scale unmanned system simulation platform containing aerodynamic modeling, and demonstrate the method effectiveness by conducting a simulation experiment with eight quadrotors in a close formation, revealing that the algorithm reduces formation tracking error by 73.9% under the influence of the aerodynamic effects.