Safe and accurate MAV Control, navigation and manipulation

Abstract

This work focuses on the problem of precise, aggressive and safe Micro Aerial Vehicle (MAV) navigation as well as deployment in applications which require physical interaction with the environment. To address these issues, we propose three different MAV model based control algorithms that rely on the concept of receding horizon control. As a starting point, we present a computationally cheap algorithm which utilizes an approximate linear model of the system around hover and is thus maximally accurate for slow reference maneuvers. Aiming at overcoming the limitations of the linear model parameterisation, we present an extension to the first controller which relies on the true nonlinear dynamics of the system. This approach, even though computationally more intense, ensures that the control model is always valid and allows tracking of full state aggressive trajectories. The last controller addresses the topic of aerial manipulation in which the versatility of aerial vehicles is combined with the manipulation capabilities of robotic arms. The proposed method relies on the formulation of a hybrid nonlinear MAV-arm model which also takes into account the effects of contact with the environment. Finally, in order to enable safe operation despite the potential loss of an actuator, we propose a supervisory algorithm which estimates the health status of each motor. We further showcase how this can be used in conjunction with the nonlinear controllers described above for fault tolerant MAV flight. While all the developed algorithms are formulated and tested using our specific MAV platforms (consisting of underactuated hexacopters for the free flight experiments, hexacopter-delta arm system for the manipulation experiments), we further discuss how these can be applied to other underactuated/overactuated MAVs and robotic arm platforms. The same applies to the fault tolerant control where we discuss different stabilisation techniques depending on the capabilities of the available hardware. Even though the primary focus of this work is on feedback control, we thoroughly describe the custom hardware platforms used for the experimental evaluation, the state estimation algorithms which provide the basis for control as well as the parameter identification required for the formulation of the various control models. We showcase all the developed algorithms in experimental scenarios designed to highlight the corresponding strengths and weaknesses as well as show that the proposed methods can run in realtime on commercially available hardware.