Disturbance Rejection Path-following Guidance Using Gaussian Process Regression via Kalman Filtering for Unmanned Aerial Vehicle

Abstract

This paper presents a disturbance rejection path-following guidance system for an unmanned aerial vehicle (UAV) based on Gaussian Process (GP) regression. The system is designed to enable the UAV to adhere precisely to predefined paths, which is crucial for effective UAV operation across various missions. In practical scenarios, environmental factors such as wind disturbances can adversely affect the tracking performance of the UAV, potentially leading to mission failure. Hence, the integration of a disturbance rejection algorithm is essential to enhance the path-following performance of the UAV. We utilize GP regression, a machine learning algorithm, as a data-driven disturbance estimator. This method has the significant advantage of requiring less prior knowledge compared to observer-based disturbance rejection algorithms. Despite its advantages, GP regression is hindered by significant computational complexity, primarily due to the necessity of inverting large matrices. A common approach to mitigate this complexity is to limit the training data size using a moving window, which discards past data. While this method reduces computational demands, it leads to the model forgetting information from past data, thereby potentially decreasing the modeling capabilities of the GP. To address these challenges, we propose integrating a path-following guidance system with GP regression through Kalman Filtering (KF). The baseline guidance system is designed using feedback linearization and optimal error dynamics. In this integrated system, the GP model is first converted into a stochastic differential equation (SDE). Subsequently, the KF is employed to estimate disturbances using the SDE model. This allows the model to be continuously updated with new data, thereby enabling the GP model to accumulate and retain information over time. The efficacy of the proposed system is validated through numerical simulations, which demonstrate both the enhanced performance of the disturbance rejection path-following algorithm and its computational efficiency. The results confirm that our approach maintains high accuracy in path-following and operates within computational constraints, making it an effective solution for real-world UAV operations.