QUADROTOR POSITION CONTROL USING ROBUST-RESILIENT CONTROLLER BASED VIRTUAL SENSOR

• Main Author: Dian Andrini - Nim: 13314014, Boby Anditio - Nim: 13314089 , Angela
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/25567
• Institution: Institut Teknologi Bandung

<p align="justify">Recent trends in control technology and robotics has led to the development of automation in flying robot technology such as quadrotor. The 6 degrees of freedom movement and nonlinear behavior, a model of a quadrotor is usually separated based on a certain range of conditions. In this research, a modelling of an inputoutput data is conducted using the Predictive Error Method (PEM) to generate a general model of a quadrotor that works under several conditions. The model will be used to design a gain parameter of the robust-resilient controller in order to control the position of quadrotor, assuming uncertainties present both in the model and its controller. This research introduced a new method in determining a quadrotor position using a virtual sensor scheme. A virtual sensor is used to determine the immeasurable variables of quadrotor using information gain from the measureable variables. The scheme consists of Diagonal Recurrent Neural Network (DRNN) which is used to make prediction of both the measureable and immeasureable variables of a quadrotor at the next time step and an Extended Kalman Filter (EKF) to correct the prediction made and estimate the immeasurable variables of quadrotor, which is the position. The DRNN is trained with an optimization algorithm, Particle Swarm Optimization, to avoid any local extreme condition in obtaining the optimal weights of the network. In this research, the proposed virtual sensor scheme integrated with a robustresilient controller has been conducted and implemented using AR Drone 2.0 in several flights. It has proven to be superior in both simulation and implementation than any previous method. <p align="justify">