Control and Motion Planning for Omnidirectional Aerial Manipulators from Free Flight to Interaction with Dynamic Objects

Abstract

With the increasing employment of robots in various industrial fields, many previously hazardous tasks have become safe and better controlled. However, there still remain a number of works that pose threats to human operators and that cannot yet be taken over by robots. Particular examples include working at height or in difficult to access or otherwise dangerous environments. Additionally, these environments can give rise to inefficiencies due to the requirements for further means, such as scaffolds, cranes, or transport by helicopter. We see aerial robots as a promising tool to assist human workers in these tasks in order to increase their safety and the viability of their work. Today, drones for both industry and end-consumers already exist in various types. Their hardware design, however, restricts them to operate in free flight, i.e., they are not designed for physical interaction with their environment. This is due to their underactuation, which does not allow to compensate for reaction forces that arise during an interaction task. Recently, more advanced aerial vehicles (Aerial Manipulators, AM) have been designed with the purpose of aerial interaction. Some of these robots are characterized by their higher degree of actuation (i.e., overactuation), which allows them to control both their pose as well as the interaction forces while being in physical contact with a structure. In this thesis, we aim to investigate different control approaches for an Overacuated Micro Aerial Vehicle (OMAV) in the pursuit of enabling manipulation of objects from the air. For that, we see the main challenges in finding optimal and safe control strategies for OMAVs, given their high degree of actuation, potential external disturbances, and fast dynamics. Further challenges lie in compensating for interaction forces and in reliable motion planning during complex manipulation tasks. We structure our investigations in a manner of increasingly complex control tasks, with the goal of developing a robot for complex aerial manipulation tasks. To this end, we first focus on the optimal control of an OMAV in free flight. Namely, we first present a PID and LQRI controller that enables an OMAV to accurately track pose trajectories while achieving secondary tasks by exploiting the platform’s overactuation. We also present a model-predictive optimal controller to execute the same task of pose tracking. In the second part we move towards the interaction with static objects. Specifically, we address push-and-slide tasks on a rigid surface with a rigid manipulator, which is a common problem for industrial inspection tasks. We show how the interaction dynamics can be modeled in an analytical way and how certain conditions can be used to distinguish between different interaction modes. Furthermore, we employ a force-torque sensor to track reference forces during the interaction. Lastly, in the third part, we focus on active manipulation of dynamic objects. We present an approach that ensures safety during the task by reducing interaction forces according to Passivity-Based Control theory. Additionally, we show a sampling-based framework that finds optimal motion policies in order to manipulate a workpiece according to a given task description. In summary, this thesis contributes to the advancement of the control of aerial manipulators and motion planning towards a higher degree of pose- and force tracking accuracy, safety, and autonomy.