Uav perception for safe flight under physical interaction
- Autores: Julián Francisco Cayero Becerra
- Directores de la Tesis: Bernardo Morcego (dir. tes.), Josep Cugueró Escofet (dir. tes.)
- Lectura: En la Universitat Politècnica de Catalunya (UPC) ( España ) en 2019
- Idioma: español
- Tribunal Calificador de la Tesis: Ramón Pérez Magramé (presid.), Manuel Vargas Villanueva (secret.), Gildas Besançon (voc.)
- Programa de doctorado: Programa de Doctorado en Automática, Robótica y Visión por la Universidad Politécnica de Catalunya
- Materias:
- Texto completo no disponible (Saber más ...)
- Resumen
The control of autonomous flying vehicles with navigation purposes is a challenging task. Complexity arises mainly due to the non-linearity and uncertainty inherently present in the flight mechanics and aircraft-air interactions. Recently, interest has grown for equipping unmanned vehicles with the capacity to interact with their environment, other vehicles or humans. This will enable interesting applications such as autonomous load carrying, aerial refueling or parcel delivering. Having measured the interaction wrenches ease the control problem which can be configured to reject disturbances or to take profit of them to fulfill mission objectives.
This thesis will contribute to this area by providing perception solutions which use limited and low cost sensors that enable state and disturbance estimation for possible, but not restricted to, interaction scenarios.
This thesis contain three parts. The first part, introduces basic concepts related to the navigation state, aircraft dynamics, and sensor models. In addition, the platform under study is presented and mathematical models associated to it are calibrated.
The second part is devoted to the observability analysis and the design of state observers. Linear and non-linear observability analysis techniques are used to unveil that the state of quadrotors equipped with GPS, magnetometers an IMU sensors cannot be uniquely identified in some specific flight configurations. Results of this section are relevant because the conflicting flight configurations contain hover, a flight maneuverer central in many unmanned aerial missions of VtoL vehicles. For many possible singular configurations, insightful descriptions and interpretations of the solution space known as indistinguishable region is provided. Findings are verified in simulation scenarios where it can be seen how a filter fails to recover the true state of an aircraft when imposing the hover flight condition. We discuss then the design of Extended Kalman Filters for state estimation that considers the available sensors. Issues that are typically not reported in the literature, such as when to update or propagate in the estimator algorithm or which coordinate frame should be used to represent each state variable are discussed. This leads to the formulation of four potentially equivalent but different discrete event-based filters for which precise algorithmic expressions are given. We compare the results of the four filters in simulation under known favorable conditions for observability. In order to diminish the effect of flying in the conflicting observability configurations, we provide an alternative filter based on the Schmidt Kalman Filter (SKF). The proposed filter shares the structure of the EKF, behaves better in the instants that the EKF fails and provides similar results in the remaining conditions.
The last part of the thesis deals with the estimation of external disturbances. Disturbance estimation results are based on the derivation of a linear model for the aircraft dynamics which then extended with a high order disturbance model to enable the estimation of fast varying disturbances. Two external disturbance estimators from the literature are reviewed and adapted to the new model. Also, two Kalman observers that exploit the linearity of the derived model are presented. A simulation comparison is provided demonstrating that the KF disturbance estimators outperform the other. In addition, this part presents a design methodology of generic quadratic bounded observers for linear systems with ellipsoidal bounded uncertainty. The derived observers maximize a user tunable compromise between the estimation convergence speed and the final volume containing the estimation error. An observer for disturbances acting on a flying platform is derived considering the high order disturbance model above mentioned. Finally, an analysis of the estimation performance with respect to the design parameters is presented.
