Master's Dissertation Defense Session No. 300 of the PPGEEC - A Practical Methodology for Fiducial Marker-Based Localization and LQR Control of a DJI Tello Quadrotor

 

STUDENT: GABRIEL DE SOUZA FERNANDES

 

DATE: August 20, 2026

 

TIME: 2:30 PM

 

LOCATION: https://conferenciaweb.rnp.br/webconf/lar-laboratorio-de-robotica-ufba

 

TITLE: A Practical Methodology for Fiducial Marker-Based Localization and LQR Control of a DJI Tello Quadrotor

 

KEYWORDS: Visual localization, quadrotor control, fiducial markers, model identification, LQI control, robust control.

 

ABSTRACT: This dissertation presents a practical methodology for localization, modeling, and control of a DJI Tello quadcopter operating in indoor environments without GPS signal. The work is motivated by the difficulty of establishing a reliable feedback loop in small commercial drones outside of ideal laboratory conditions, where external motion capture systems may not be available and visual perception is affected by noise, occlusions, limited field of view, and communication delay. Instead of proposing a new control theory, the dissertation integrates established techniques into a reproducible experimental workflow for real-world flight testing with a low-cost quadcopter. The methodology combines fiducial marker-based localization, data-driven model identification, and LQR control with integral action. The localization system utilizes a calibrated camera model, a known marker map, pose estimation with multiple markers, measurement validation, and Kalman filtering. Static tests with different marker visibility configurations demonstrate the importance of marker geometry and justify discarding poorly conditioned pose measurements when few markers are visible. The dynamic model is identified from experimental data, using a reduced representation of velocity commands that relates high-level commands to translational and yaw response. A multi-objective NSGA-II procedure is employed to select parameters that balance position and velocity adjustment, while the resulting Pareto fronts provide uncertainty intervals for the design of a robust controller. An LQR formulation with integral action, called LQI, is then synthesized from the identified model, taking into account parametric uncertainty. The controller is applied through a virtual control formulation that maps the LQI input to the limited command interface of the DJI Tello. Real-world flight experiments, using visual localization as feedback, are performed for tracking circular trajectories and waypoints, without external ground truth reference during final control tests. The results indicate stable and repeatable behavior under practical sensing and actuation constraints, also revealing limitations associated with the quality of visual estimation, apparent delay, axis coupling, command saturation, and the simplified model structure. In short, the dissertation contributes an experimentally grounded procedure for the application of fiducial marker localization, data-driven identification, and robust LQR/LQI control in a real quadcopter.

 

EXAMINING COMMITTEE MEMBERS:

 

ANDRE GUSTAVO SCOLARI CONCEIÇÃO (ADVISOR)  - UFBA (Chair)

 

HUMBERTO XAVIER DE ARAUJO (CO-ADVISOR)        - UFBA

 

TIAGO TRINDADE RIBEIRO                                     - UFBA

 

TITO LUIS MAIA SANTOS                                        - UFBA

 

ANDRE LUIS MARQUES MARCATO                            - UFJF

Em 18/08/2026