1- Sharif University of Technology
Abstract: (897 Views)
In this paper, a model-free nonlinear architecture is presented to control a fixed-wing UAV. This architecture has inner and outer loops. The inner loops, which are designed based on convolutional neural networks, control the internal dynamics of the aircraft in a model-free procedure. The outer loops, which use conventional linear controllers, are designed to control the kinematics of the UAV. The neural networks used to control the inner loops are trained offline based on generated databases to avoid time-consuming online learning processes. These databases are created by simulating simple training models. Then, the input-output data of these training models are pre-processed and mapped to image frames so that they can be given as input to convolutional neural networks. After that, a suitable network structure is selected and the networks are trained based on the mapped databases. These trained networks, together with cascaded linear controllers, are applied to the nonlinear simulation of the fixed-wing UAV and its performance is investigated. The inner loops that control the internal dynamics of the UAV have been applied in both single-stage and two-stage forms and their performance has been compared. Also, the sensitivity of the controller to measurement noises and disturbances has been investigated.
Type of Article: Research paper |
Subject:
Special
Received: 2023/05/29 | Accepted: 2023/12/18 | ePublished ahead of print: 2024/01/1
Received: 2023/05/29 | Accepted: 2023/12/18 | ePublished ahead of print: 2024/01/1
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
