A UAV Dynamics Model Based on Machine Learning

Abstract

The developed machine learning based drone dynamic model works at a fraction of real-time with 2.3GHz CPU for a given recorded radio controller data set of a figure of eight trial. The high-level shape comparison between the actual pattern and predicted pattern is at a satisfactory level. However the shape of the predicted path is closer to the shape of a lace eight and there are many predicted points are not in the required figure of eight pattern. It may due to the data preprocessing techniques, minute effect of the wind or the used machine learning technique. Carried out experiments provide encouraging results to further experiments and validations to clarify the impacts of deploying Figure of eight maneuvering pattern or its derivative to collect data and built machine learning based drone dynamic model.