DEVELOPMENT OF QUADCOPTER TRAJECTORY AND ATTITUDE CONTROL BASED ON VISUAL FEEDBACK USING VIRTUAL ENVIRONMENT

• Main Author: Ardy Putra, Iqbal
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/75221
• Institution: Institut Teknologi Bandung

The visual inspection is an integral part in the aircraft’s Maintenance, Repair, and Overhaul (MRO) industries. Alas, the current practice is still conducted manually by human and many studies see this as inefficient and ineffective process. The prominent approach to improve this practice is by utilizing robot as an inspection platform and one of the main challenges is that it needs precise localization sensor, especially in indoor environment where the inspection is conducted, so that the platform could be maintained at predefined trajectory without harming the aircraft. In this nuance, we will use quadcopter as the inspection platform since it has the ability to cover the whole aircraft and the localization system will use stereo vision as the sensor, neglecting the usual use of GPS and IMU due to its limitation on signal availability and accuracy issue. This research generally attempts to evaluate the performance of the stereo vision, its integration with the quadcopter, and the overall system performance in several trajectories. The research is mainly conducted in numerical simulation and by utilizing the advantages of virtual environment in order to model the vision sensor realistically in ideal conditions. The development of such system is conducted using MATLAB/Simulink to model the quadcopter dynamics paired with Gazebo Simulator to create a virtual environment and model the camera sensor. The generated image data is segmented using color thresholder and blob analysis in order to detect the markers on the quadcopter thus indirectly localize the platform. The result shows that the reconstruction algorithm, with maximum camera’s height variation at 7.5 m, gives reliable and stable estimation in the order of centimeter accuracy, however the mean of error generally increases as the camera’s height escalates. Also, the performance of X and Y position estimation decreases as the quadcopter moves to near camera’s FoV limit. In addition, the overall system performance, which consists of quadcopter, stereo vision, and control system using LQR, shows great performance in trajectory tracking without heading. Even so, the system still fails when conducting simulation together with heading variation.