Reinforcement learning-based controller design for a ‎proposed octorotor with tilt-arm angles ‎

Document Type : Research Article

Authors

1 Department of Mechanical Engineering, Faculty of Engineering, University of Isfahan,

2 Department of Mechanical Engineering, University of Kashan, Kashan, Iran

3 Department of Mechanical Engineering, Kashan University

4 Department of Mechanical Engineering, Faculty of Engineering, University of Isfahan, Isfahan, Iran

Abstract

The maneuverability of a quadrotor or octorotor UAV is limited in the standard configuration ‎because the force vectors of the propellers are parallel and only have four active degrees of ‎freedom. Therefore, they lack the controllability of six independent degrees of freedom. This ‎study designs a novel configuration for an octorotor capable of hovering with roll or pitch angles in ‎a specific position, contrary to UAVs with a standard configuration that can only hover in a ‎horizontal position. In other words, in this octorotor, orientation tracking is also added to the ‎octorotor's targets in addition to position tracking. The proposed model can be controlled by ‎altering the velocity of the eight rotors and the tilt angle of the four arms. Such alterations in ‎velocity and tilt angle are such that they can provide the aerial vehicle with most optimum ‎maneuverability. After deriving the proposed dynamic octorotor model, a controller is proposed ‎using neural networks (NNs) and reinforcement learning (RL), capable of controlling the proposed ‎octorotor with six independent degrees of freedom. Finally, trajectory tracking, octorotor position, ‎and controller robustness to possible motor malfunctions are examined, and numerical simulation ‎results are provided.‎

Keywords

Main Subjects

Amirkabir Journal of Mechanical Engineering
Volume 55, Issue 10
January 2024
Pages 1175-1194

Files

Share

How to cite

Statistics