Methods for Landing, Position Estimation, and Recovery for a Single and Multi-UAV System Transporting a Payload

This doctorate research project considers a system composed of multiple Unmanned Aerial Vehicles (UAVs) transporting a payload through flexible cables. Supposing the payload has a not negligible weight or is not carriable by a single drone without compromising the flight stability, a scenario of a UAV having a failure during transportation and automatically detaching from the system is considered, leading to the development of recovery strategies. This means that the failing UAV would suddenly stop contributing to achieving the goal, and it would be detached from the team so that it does not become part of the payload itself and does not contribute to undesired behaviors. The dynamics describing the flight behavior of the system with multiple quadrotors are described by a set of non-linear equations, while the tensions acting on the rods are modeled as a mass-spring-damper system. An algorithm based on a Nonlinear Least Squares regression and Gauss-Newton steepest descend procedure is adapted and developed to optimize the value of the parameters in the proposed model. To restore the stability to the remaining team in the considered scenario, a recovery strategy is proposed, using a combination of control algorithms such as Proportional, Integrative and Derivative control, Integrative Sliding Mode Control, Consensus-Based control, and formation change procedures. The problem of mutual localization among robots and landing is also addressed, and the case of one quadrotor transporting a light payload and landing on a catamaran in a marine environment is considered. Localization methods based on perception are presented, which exploit the use of cameras on the quadrotor and the presence of several AprilTags placed on a platform on the catamaran. Algorithms for the landing on the catamaran are introduced through the use of the Apriltags and an ultrasonic sensor, which helps the procedure when the distance with the catamaran is close. The considered software and hardware architecture is presented. Software-In-The-Loop tests are performed to test the recovery methodology on a system composed of three UAVs carrying a payload and on a single UAV landing on a simulated catamaran, to validate the landing and position estimation techniques. In particular, simulations with three UAVs and the payload are performed with multiple goals: first, comparing the performances of two different variations of the considered control law in a disturbance-free environment, including also a formation change technique that helps to mitigate the oscillations of the payload; second, gathering the system state and tension data after the failed quadrotor detaches and use it to identify the parameters of the tension model with the Nonlinear Least Squares algorithm. Third, to validate the proposed control law in the presence of wind and gust disturbances. Considering the remaining two UAVs after the failure and the detachment of one quadrotor, a brief study of the static properties of such a system is performed, by analyzing the limits of the configuration of the formation. Simulations with one quadrotor are then used to validate the proposed landing and position estimation techniques, by relying on data recorded from an experiment with a catamaran in a marine environment. Further experiments with the real quadrotor have been performed to gather vision data, used subsequently to confirm the performances of the position estimation technique. Finally, considerations about the system performance are made, along with possible implementations for future studies.