Search

Over-actuated MAVs have received growing attention in recent years as they allow for physical interaction with their environment. Most interaction tasks so far have been conducted in controlled laboratory environments with little to no outside disturbances. To perform high-accuracy interaction tasks in outdoor conditions, we need a method to reliably distinguish wrenches (i.e. forces and torques) that arise from interaction at the contact point from those that arise from external disturbances (such as aerodynamic effects from wind). The applicable methods to achieve this depend heavily on the used sensor equipment. Therefore, we propose to use a combination of odometry and force/torque sensors, together with an analytical model of the platform dynamics. Thus we can estimate contact and disturbance wrenches simultaneously. In this project, we present a model-based filtering method, namely Extended Kalman Filter (EKF). Following this, an estimation framework is designed and implemented in simulation and verified on the physical platform. With a combination of a novel (especially for MAVs) hybrid motion-force controller, we enable physical interaction of such vehicles with the environment under the presence of external disturbance. For a contact-based inspection, we provide the platform with two types of manipulators. Initially, we validate our approach with a simple rigid manipulator and subsequently with a delta arm attached to the vehicle. We prove the ability of the entire control framework in real-world experiments. Thus, we verify the quality of the estimates by comparing them to ground-truth measurements and showcasing possible industrial applications.

This doctoral thesis addresses the study of omnidirectional tilt-rotor aerial robots, and their application to aerial physical interaction tasks. Through modelling, control, prototype design, and experimental evaluation, this work carves a new direction in aerial robotics research, and seeks to inspire a future of versatile and autonomous aerial manipulators. Recent developments in the field of fully actuated aerial robots have demonstrated the exceptional advantages of these systems for physical interaction. Characterised by their decoupled translational and rotational system dynamics, these systems not only outperform their underactuated counterparts, but extend their capabilities. Through the dynamic re-orientation of actuated thrust vectors, we now have access to a great expanse of possible morphologies, dynamic system capabilities, and new applications. Extending these novel tilt-rotor systems with an active manipulator further demonstrates enhanced end effector performance for manipulation tasks. The concept of macro-micro manipulation -- using a highly dynamic end effector mounted to a powerful base -- overcomes dynamic limitations that currently restrict the efficacy of aerial manipulators. In pursuit of versatile and high performance systems for aerial physical interaction, the present work combines these concepts to advance the state-of-the-art in aerial manipulation. The design space of a tilt-rotor aerial robot is selected by optimizing a general model around desired performance metrics and system parameters. The resulting system, chosen for a balance of omnidirectional and efficient flight capabilities, is compared against other state-of-the-art fully actuated systems. Aerial interaction models are developed for fixed and active manipulators, and a geometric optimization is performed to determine the design of a parallel manipulator in the context of an omnidirectional flying base. The control problem divides the system conceptually into tracking control of a pure wrench generating base, and a subsequent actuator allocation problem to achieve a six degrees of freedom wrench with 18 individual actuator commands. The nonlinear and highly dimensional actuator space is addressed with instantaneous and differential allocation methods, the latter of which incorporates secondary tasks, such as the unwinding of tilt-arm cables, in the actuation null space. Inverse-dynamics based controllers are introduced for control of the flying base, treating the whole tilt-rotor system as a single rigid body. Interaction controllers including axis-selective impedance and direct force control are developed for the system equipped with a fixed manipulator arm. A redundant control strategy is developed for the omnidirectional system with an attached translational parallel manipulator, where predicted reaction forces are fed to an independent base controller to compensate the manipulator dynamics. Several iterations of omnidirectional tilt-rotor aerial robots are designed and constructed, considering the requirements of aerial interaction tasks. Actuator selection, tilt-rotor mechanisms, and complete system assembly are presented, as well as design details for a parallel manipulator. Experimental trials evaluate the capabilities of the physical system and its control implementation to track omnidirectional trajectories. Aerial physical interaction tasks are demonstrated, involving point force application with the environment, push-and-slide tasks, and applications to non-destructive contact inspection of concrete. Fast end effector tracking and disturbance rejection experiments are performed to validate the macro-micro concept of an omnidirectional tilt-rotor parallel manipulator. Ranging from general modelling to control, design choices and complete system prototypes, the content of this work acts as a guide for envisioning and building innovative systems that will push the frontier of aerial manipulation.

Cooperative robotic problems often require coordination in space in order to complete a given task, important examples include search and rescue, operations in hazardous environments, and autonomous taxi deployment. Events can be quickly detected by partitioning the working environment and assigning one robot to each partition. However, a crucial factor that limits the effectiveness and usage of coverage algorithms is related to the ability of taking decisions in the presence of constraints. In this paper, we propose a coverage control algorithm that is capable of handling nonlinear dynamics, and state and input constraints. The proposed algorithm is based on a nonlinear tracking model predictive controller and is proven to converge to a centroidal Voronoi configuration. We also introduce a procedure to design the terminal ingredients of the model predictive controller. The effectiveness of the algorithm is then highlighted with a numerical simulation., IFAC-PapersOnLine, 53 (2), ISSN:2405-8963, 21th IFAC World Congress. Proceedings

In addition to geometrical and thermal, mainly dynamic errors occur on mechatronic systems such as machine tools. The dynamic errors are of transient type and caused by forces, which are acting on the structural components of the mechatronic systems during motion, interacting with a motion controller reacting to them. Besides static and transient internal and external disturbances, mainly inertial forces lead to deformations of the compliant structural components, causing deviations at the tool center point (TCP). Especially for high dynamic systems with relatively soft and therefore compliant structural components, which are moved at high accelerations, the dynamic errors become dominant. Mainly by restrictions related to the applied measurement principle, these errors are not fully detectable by the axes measurement systems. Often linear encoders are used for measuring the locations of the moving structural components, which are forced to be placed far from the TCP due to limitations in design. Accordingly all dynamic deviations in-between the measuring locations of the encoder and TCP are invisible. Within this work the available sensor principles and their usability for measuring the dynamic deviations between encoder and TCP are investigated. It is shown that a sensor fusion approach, combining the linear encoder with additional accelerometer measurements at the workpiece (WP) and the TCP leads to a significant reduction of the dynamic measuring error between encoder and TCP. For the sensor fusion task a linear filter based method is presented, which combines the encoder with the accelerometer measurements with a quasi-static deformation model. Further a parametrisation strategy for the filtering task is derived based on a simplified model of the dynamic error between WP and TCP. Using simulations and measurements on a validated multi-axes mechatronic system it is shown that the dynamic error can be measured with an accuracy of at least 10 um for highly dynamic systems. Further an estimation error calculation is derived. As nowadays reasonably accurate accelerometers are available at low costs on the market, the method might be broadly applied on machine tools, in order to enhance the measurement accuracy. In addition to the measurement of the dynamic error, the reduction of it is the main objective, which might be achieved by either online or offline dynamic compensation methods. Offline dynamic compensation methods often rely on accurate measurements of the actual deviations at the TCP . These algorithms might directly benefit from the increased measurement accuracy of the method described above. As a first approach for online dynamic compensation, within this work a new parametrisation strategy for the standard cascaded loop motion controller is presented, which is minimising the dynamic error at the TCP. It considers the dynamics between encoder and TCP in order to compensate for quasi-static deviations and maximising the structural damping actively at the same time. The parametrisation strategy is validated using measurements and simulations. As it is applicable to almost every industrial numerical control (NC), it might directly be used for the dynamic performance optimisation of industrial machine tools. The second presented approach involves an additional velocity feedback loop, added to the standard cascaded loop motion controller. The known principle in literature is applied to machine tool direct drive structures including the derivation of a hands-on parametrisation strategy. Using the additional velocity feedback allows to increase the controller bandwidth significantly without loosing structural damping. By this the disturbance rejection and tracking accuracy is increased without induction of unwanted vibrations. The method is applied to simulations, showing promising results. For the practical applicability of the approach online measurements of the velocity at the TCP are needed, where the accelerometer based measuring method is suggested to be extended from offline to online applicability., ISBN:978-3-907234-86-0

IEEE Transactions on Robotics, 31 (6), ISSN:1552-3098, ISSN:1042-296X, ISSN:1941-0468

ISBN:978-3-906031-72-9

IMAV 2014 : International Micro Air Vehicle Conference and Competition 2014

IEEE International Conference on Robotics and Automation (ICRA), 2013 : 6 - 10 May 2013, Karlsruhe, Germany, ISBN:978-1-4673-5640-4, ISBN:978-1-4673-5643-5, ISBN:978-1-4673-5641-1

In this paper, we present our latest achievements towards the goal of autonomous flights of an MAV in unknown environments, only having a monocular camera as exteroceptive sensor. As MAVs are highly agile, it is not sufficient to directly use the visual input for position control at the framerates that can be achieved with small onboard computers. Our contributions in this work are twofold. First, we present a solution to overcome the issue of having a low frequent onboard visual pose update versus the high agility of an MAV. This is solved by filtering visual information with inputs from inertial sensors. Second, as our system is based on monocular vision, we present a solution to estimate the metric visual scale aid of an air pressure sensor. All computation is running onboard and is tightly integrated on the MAV to avoid jitter and latencies. This framework enables stable flights indoors and outdoors even under windy conditions.

2011 IEEE/RSJ International Conference on Intelligent Robots and Systems : (IROS 2011) ; San Francisco, California, USA, 25 - 30 September 2011, ISBN:978-1-612-84454-1, ISBN:978-1-612-84455-8

The goal of this thesis is to develop novel model-based methods for optimal control and optimal design of hybrid electric vehicles. Two different approaches are used when designing the energy management strategy for two types of parallel hybrid electric vehicles. For the full parallel hybrid a pure mathematical approach is used while an approach derived from optimal sizing studies is used to design the energy management strategy for the torque-assist parallel hybrid. The optimal control problem associated with the energy management in a full parallel hybrid is solved explicitly for a simplified model. The solution of the optimal control problem shows how optimal energy management strategies are derived and that the solution yields simple rules depending on vehicle parameters. Furthermore, a causal, real-time control strategy including anti-windup is presented. The novel energy management for the torque-assist hybrid shows that the gear shifting control can be separated from the torque split control. The energy management strategy utilizes the gear shifting strategy for control of the energy flows while the torque split strategy is given by a simple rule. Results show that the proposed energy management strategy achieves a fuel consumption within 1% from the global optimum for most driving cycles. Furthermore, the results are not sensitive to limitations and energy losses associated with gear shifting. Both the full and the torque-assist parallel hybrid vehicles are optimized with respect to the component dimensions. The overall power-to-weight ratio is kept constant while the hybridization ratio is optimized and investigated for the full hybrid and the torque-assist hybrid. The study shows the non-intuitive result that the need for hybridization is larger in the torque-assist hybrid than in the full hybrid. The simplicity of the torque-assist hybrid allows the optimal hybridization ratio to be found using a very simple and computationally cheap rule. The objective of this rule is to minimize the total CO2 emissions of the vehicle, while maintaining its drivability at a constant level. The starting point is an analysis in which the optimal energy management strategy is found for eight typical driving cycles using dynamic programming. Analyzing these results, a simple yet powerful rule-based method is proposed that allows choosing the sizes of the combustion engine and of the electric motor such that the CO2 emissions are very close to the minimum value, i.e., with a deviation of less than 1% for most driving cycles. In the last chapter of this thesis the focus is on the dynamic programming algorithm. Issues related to the implementation of the dynamic programming algorithm for optimal control of a one-dimensional dynamic model is investigated. A study on the resolution of the discretized state space emphasizes the need for careful implementation. A novel method is presented to treat numerical issues appropriately. In particular, the method deals with numerical problems that arise due to high gradients in the optimal cost-to-go function. These gradients mainly occur on the border of the feasible state region. The proposed method not only enhances the accuracy of the solution but also allows for a reduction of the state-space resolution with maintained accuracy. The latter substantially reduces the computational effort to calculate the global optimum. Finally, the improved dynamic programming algorithm is summarized and implemented in a general, public Matlab function. Das Ziel dieser Arbeit ist die Entwicklung neuer, modellbasierter Methoden fur die optimale Regelung und Auslegung von elektrischen Hybridfahrzeugen. Zwei verschiedene Ansatze werden angewandt fur die Entwicklung des Energiemanagements fur zwei Typen von parallelen elektrischen Hybridfahrzeugen. Fur den Fall eines Vollhybridfahrzeuges wird ein rein mathematischer Ansatz angewandt, wahrend das Energiemanagement fur ein mildes Hybridfahrzeug aus Studien der optimalen Auslegung abgeleitet wird. Das Optimal Control Problem fur das Energiemanagement von parallelen Vollhybridfahrzeugen wird explizit gelost fur ein vereinfachtes Modell. Diese Losung zeigt auf, wie optimale Energiemanagementstrategien hergeleitet werden und dass die resultierenden Losungen einfache Regeln darstellen, welche durch die Fahrzeugparameter definiert sind. Desweiteren wird eine darauf basierende kausale Echtzeitstrategiemit Anti-Windup hergeleitet. Ein neuartiges Energiemanagement fur milde Hybridfahrzeuge zeigt auf, dass die Gangwahl von der Regelung der Momentenaufteilung separiert werden kann. Dieses Energiemanagement verwendet die Gangwahl zur Steuerung der Energieflusse wahrend die Momentenaufteilung durch einfache Regeln vorgegeben wird. Die Resultate zeigen, dass dieses Energiemanagement Treibstoffverbrauche erzielt, welche fur die meisten Fahrzyklen weniger als 1% Abweichung vom globalen Optimum aufweisen. Die erzielten Resultate sind insensitiv bezuglich Beschrankungen und energetischen Verlusten bei Gangschaltvorgangen. Sowohl das Voll- wie auch das Mildhyridfahrzeug werden bezuglich Komponentendimensionierung optimiert. Das Leistungs/Gewichts-Verhaltnis wird konstant gehalten wahrend der Hybridisierungsgrad optimiert wird. Der Unterschied zwischen optimalem Hybridisierungsgrad fur Vollbeziehungsweise Mildhybridfahrzeug wird untersucht. Die Resultate zeigen uberaschenderweise, dass milde Hybridfahrzeuge hohere optimale Hybridisierungsgrade benotigen als Vollhybridfahrzeuge. Die Einfachheit von milden Hybriden erlaubt den optimalen Hybridisierungsgradmittels einer sehr einfachen und rechnerisch effizienten Regel zu finden. Das Ziel dieser Regel ist die Minimierung der totalen CO2 Emissionen des Fahrzeuges, wobei die Fahrleistungen konstant gehalten werden. Ausgangspunkt ist eine Analyse, in welcher die optimale Energiemanagementstrategie mittels Dynamischer Programmierung fur acht typische Fahrzyklen ausgewertet wird. Die Untersuchung dieser Resultate erlaubt die Herleitung einer einfachen, aber effektiven Methode, die es erlaubt, die Dimensionierung des Verbrennungsmotors und des Elektromotors so zu wahlen, dass die CO2 Emissionen sehr nahe beim Minimum liegen. Die Abweichungen sind kleiner als 1% fur die meisten Fahrzyklen. Das letzte Kapitel dieser Arbeit fokussiert auf den Algorithmus der Dynamischen Programmierung. Probleme bezuglich der Implementierung der Dynamischen Programmierung fur eindimensionale Optimal Control Probleme werden untersucht. Eine Studie der Auflosung des diskretisierten Zustandsraumes verdeutlicht die Notwendigkeit einer exakten Implementierung. Eine neue Methode wird vorgestellt, die es erlaubt, diese numerischen Probleme korrekt zu behandeln. Die Methode behandelt numerische Probleme, die aufgrund grosser Gradienten in der optimalen Restkostenfunktion auftreten. Diese Gradienten treten meist am Rand des losbaren Zustandsraumes auf. Die vorgestellte Methode verbessert nicht nur die Genauigkeit des gefundenen Losungen, sondern sie erlaubt auch die Auflosung des Zustandsraumes zu reduzieren bei gleicher Genauigkeit wie im ursprunglichen Problem. Diese Reduktion der Auflosung reduziert den Rechenaufwand fur die Auswertung des globalen Optimums wesentlich. Der verbesserte Algorithmus der Dynamischen Programmierung wird letztlich zusammengefasst und in einer Matlabfunktion implementiert.

ISBN:978-3-906467-75-7

Fortschritt-Berichte VDI. Reihe 8, Mess-, Steuerungs- und Regelungstechnik, 1132, ISBN:978-3-18-513208-7, ISBN:3-18-513208-4

ISBN:3-86537-905-2

ISBN:3-9522686-9-0

ISBN:3-906483-05-3

ISBN:3-906483-04-5

ISBN:3-260-05406-5