Adaptive Neuro-Fuzzy Inference System identification for the dynamics of the AR.Drone Quadcopter

Abstract

In this paper, we study the non-linear modelling of the lateral, longitudinal, and vertical dynamics of the AR.Drone Quadcopter by means of the Adaptive Neuro-Fuzzy Inference Systems (ANFIS) identification technique. We derive the multi-input multi-output (MIMO) ANFIS model of the inner (attitude) loop systems. We employ the Dryden wind turbulence model to represent a realistic wind gust scenario in real flight environments. Furthermore, we benchmark the performance of our proposed ANFIS models with respect to the performance of the linear system identification techniques. This paper serves as a preliminary study towards our long-term goal in developing robust autopilot systems for a quadcopter drone using neuro-fuzzy models, which have multiple advantages over the traditional (mathematical-based) modelling techniques, such as the knowledge-based property of the fuzzy principle, which is not only transparent, but also suitable to accommodate ambiguity; in addition to its adaptive nature. The learning capability of the neural networks is also suitable to represent the dynamics of highly non-linear systems as in the case of our quadcopter drone.