Orientation estimating in a non-modeled aerial vehicle using inertial sensor fusion and machine learning techniques

Abstract

Unmanned Aerial Vehicles (UAV) have offered alternatives for applications in which human integrity is compromised. In this sense, the need of increasing autonomy in these vehicles presents an alternative to artificial intelligence areas to enhance navigation capacities through several environments. This article presents an evaluation for estimating inclination and orientation, using automatic learning algorithms for a dynamic multi-rotor plant. To do so, an experiment is proposed to collect the data from multiple IMU sensors over an UAV main board, and submitted to different inclinations before achieving the classification task. The reported results using k nearest neighbors (k − NN), support vector machines (SVM) and Bayes show eficiency during the recognition, obtaining an accuracy score up to 99 %. Besides, the algoritms could be combined along with robust control techniques, which is ideal for implementation in embedded systems with low processing capacities.