Over the past few decades, research in the field of Quadrotor control has gone through a significant alteration. The centre of focus moved from the piloted UAVs to the autonomously guided vehicles as there exist numerous situations, where there is no possibility, time or pilot for a man controlled Quadrotor. It is an even bigger challenge to accomplish autonomous navigation in indoor environments, where there are tight constraints for physical boundaries and need higher precision pose estimation then GPS measurements could provide.
This work has a dual purpose, to develop a system, where miniature quadrotors are able to perform a real-life indoor UAV racetrack autonomously with the fastest and most robust performance. Furthermore, to create a tesbed, where researchers and students are able to test their control algorithms on real hardware, where easy implementation and state estimation is given. To reach the goals, the core of the intelligent space, the motion capture system needed to be set up, and included into the higher level system with the host PC. Here a ROS environment was running on the PC maintaining the connection and the simultaneous communication between the hardware and sensor peripherals including the Crazyflie itself.
On top of a working state estimation and framework to handle communication, the control methods were tested for the best performance autonomous flights. For the Quadrotor control, first an off-board cascade linear PID control method was implemented and designed for hover state navigation, only suitable for small angle excursions. As a more aggressive flight performance was desired for UAV races, with the use of the full dynamical capability of the Crazyflie, later another nonlinear control method was implemented, based on 3D rotation matrix representation, with fully specified differential equation of the dynamics in the function of the differentially flat in- and outputs. This control law was suitable for even very large angle orientation changes, thus resulting aggressive acceleration performance in the horizontal plane as well.
Here, for the full autonomous flights, the last task was to implement the most suitable trajectory generation method for optimized UAV trajectories. To fulfill this, we needed to take into account the trajectory time optimization, for the fastest racetrack performance, and the minimum possible departure of the generated trajectory from the nominal path, for obstacle and wall avoidance. Finally, optimization for computational time to reserve the possibility, to later incorporate the method into the on-board firmware, running on microcontrollers with limited capability. After studying and testing several methods on the real hardware, the most suitable was chosen and put into practice with cost functions including the time variable, and any of the position vector derivatives, allowing optimization for the snap of the quadrotor, for the best battery and actuator management.
To summarize, the goal of the work was reached with creating the MTA SZTAKI MIMO arena (MIcro aerial vehicle and MOtion capture), a working autonomous UAV racetrack for high-speed navigation, and a working real hardware testbed that enables easy feasibility for any kind of developed control and estimation method for researchers and educational purposes.