Refinement of UAV dynamics model through online identification: a model-data hybrid approach

This paper investigates how to precisely and efficiently obtain the dynamics model of Unmanned Aerial Vehicle (UAV) which can used in single or swarm control. The UAVs are homogeneous in the mechanical structure, yet differ in a variety of unmodeled uncertainties, which cannot be ignored in high-fidelity simulation systems. To solve this, we propose a model-data hybrid approach that consists of two steps: 1) establishing a baseline model by first-principle; and 2) refining the model through online identification. More specifically, the baseline model is a linearized state equation with pending parameters. The parameters are adaptively determined through a least-squares identification algorithm based on real-time flight data. Finally, we design a comprehensive flight mission with 5 common modes (climbing, level flighting, turning, descending, and circling) for verification. The results of simulation experiment present a good consistency between the outputs of the refined dynamics model and the real flight data.

I. Introduction

Currently, the development of Unmanned Aerial Vehicle (UAV) is gradually transforming from single-platform to multi-platform. Hence the challenges of UAV swarm control emerge : how to provide a control method for ground control station when the scale of the UAV swarm expands faster than the speed of UAV's autonomous capability improvement, and how to effectively deal with various accidents while reducing the burden of operators^[1].