Fully autonomous flocking behavior of flapping wing robots

The present work proposes a solution for performing indoor flocking with micro air vehicles (MAVs) in a fully autonomous way, considering the motion of the robots in three dimensions. The main characteristics that are looked for in the behavior of the flock are: adaptability, scalability and robustness. In order to meet these goals, each agent is equipped with on-board sensors that allow it to navigate autonomously in the environment; specifically, the IMU provides information about the velocity and acceleration of the drone and about its pose, while an UWB antenna measures the relative distance from every other agent in the flock and broadcasts the information extracted by the IMU to the rest of the swarm. Basing on these measurements only, a state estimator allows each agent to obtain accurate relative localization of any other agent within UWB range of work. The estimation is done by means of an Extended Kalman Filter, which outputs both the relative position and the relative yaw angle of a MAVj (tracked) with respect to a MAVi (host). The resulting system does not rely on any external infrastructure, such as UWB beacons or the GNNS, making the flock deployable anywhere at any time. The limitations of this solution are studied through an observability analysis in Lie derivatives, which leads to the identification of conditions in terms of combinations of inputs and states, that would affect the convergence of the filter. Finally, two flocking algorithms are presented, inspired by the Reynolds' theory of boids' rules of alignment, cohesion and separation. Emphasis is given to the implementation of a completely decentralized control, such that the resulting system appears highly robust to the loss of one or more agent and at the same time scalable to changes in the size of the flock, whose members can be easily added or removed without changing anything in the algorithm. The overall collective motion that emerges is the result of purely local interaction, without the need for a preprogrammed path or formation control. This feature makes the flock indipendent from human guidance and able to adapt to any environment or situation dynamically. The EKF and the flocking algorithms are coded in MATLAB & Simulink in discrete time and integrated with the preexisting model of a flapping wing MAV, the DelFly Nimble, designed and developed at the Technical University of Delft.