In this work we propose a number of relevant engineering applications that exhibit both a continuous and a discrete evolution, and are therefore suitably described by a recent formalism for hybrid dynamical systems. More specifically, (i) we design observer schemes for a nonsmooth disturbance arising in AC/DC conversion, which we then cancel from a desirable signal; (ii) we show how reset actuation applied to nonlinear mechanical systems can at the same time sustain or damp oscillations; (iii) we study the feedback interconnection of a classical proportional-integral-derivative controller with a sliding mass under Coulomb friction through differential inclusions. In the context of dynamical systems, we analyze the properties of these applications in terms of asymptotic stability through Lyapunov functions tailored for hybrid systems. Instead of the standard Lyapunov conditions, we prove asymptotic stability through weaker, or relaxed, conditions that are compensated by additional (structural) properties that may be easier to verify.

In this thesis we study linear time-invariant systems feedback interconnected with three specific nonlinear blocks; a play/stop operator, a switching-reset mechanism, and an adaptive dead-zone. This setup resembles the Lure problem studied in the absolute stability framework, but the types of nonlinearities considered here do not satisfy (in general) a sector condition. These nonlinear blocks give rise to a whole range of interesting phenomena, such as compact sets of equilibria, hybrid omega-limit sets, and state constraints. Throughout the thesis, we use the hybrid systems formalism to describe these phenomena and to analyze these loops. We obtain sharp stability conditions that can be formulated as linear matrix inequalities, thus verifiable with numerically efficient solvers. Finally, we apply the theoretical findings to two automotive applications.

Model Predictive Control (MPC) is the de facto standard in advanced industrial automation systems. There are two main formulations of the MPC algorithm: an implicit one and an explicit MPC one. The first requires an optimization problem to be solved on-line, which is the main limitation when dealing with hard real-time applications. As the implicit MPC algorithm cannot be guaran- teed in terms of execution time, in many applications the explicit MPC solution is preferable. In order to deal with systems integrating mixed logic and dynam- ics, the class of the hybrid and piecewise affine models (PWA) were introduced and tackled by the explicit MPC strategy. However, the resulting controller complexity leads to a requirement on the CPU/memory combination which is as strict as the number of states, inputs and outputs increases. To reduce drasti- cally the complexity of the explicit controller while preserving the controller’s performance, a strategy combining switched MPC with discontinuous simpli- cial PWA models is introduced in this thesis. The latter is proven to be circuit implementable, e.g., in FPGA. To ensure that closed-loop stability properties are guaranteed, a stability analysis tool is proposed which exploits suitable and possibly discontinuous PWA Lyapunov-like functions. The tool requires solving offline a linear programming problem. Moreover, the tool is able to compute an invariant set for the closed-loop system, as well as ultimate boundedness and input-to-state stability properties.

Autonomous driving is an important research field and presents several problems with different nature and complexity. The goal of this work is describing such problems and the concerning solutions adopted while developing an experimental vehicle designed to autonomously perform some manoeuvres and/or to be remotely driven: the vehicle RUMBy. The problem has been faced from its beginning, i.e. from the hardware and software design: the requisites the system should satisfy are discussed, such as the vehicle and the sensors chosen, and the adopted hardware and software architectures are then described. Some mathematical models representing the vehicle dynamics are also presented, that have been employed in several applications, from the dynamic behaviour simulation up to the control system synthesis. Each model besides presents characteristic parameters that should be evaluated: therefore the problem of identification is discussed in detail, accompanied by the results obtained during an experimental activity carried out in collaboration with a Japanese research institute. Also the vehicle state estimation constitutes a key point in autonomous driving field: about this, the experimental results yielded by an estimation algorithm, based on a Kalman filter, are presented and discussed. Finally, the yaw rate control problem is examined, which is fundamental for both the motion stabilization (during remote driving) and for the following of yaw rate reference profiles (while performing autonomous manoeuvres). Two control architectures, based on a disturbance observer, have been developed and compared in a simulation campaign, that has been carried out by means of a static simulator which reproduces the driving of RUMBy in a virtual environment. This work aims then at marking a milestone within a work in progress, as well as at representing a potential guideline for researchers that would coping with projects concerning autonomous driving.

Haptic devices are becoming a common and significant tool in the perspective of robotic neurorehabilitation for motor learning, particularly in post-stroke patients. As a standard approach, this kind of devices are used in a local environment, where the patient interacts with a virtual environment recreated in the computer's screen. In this sense, a general framework for virtual reality based rehabilitation was developed. All the features of the framework, such as the control loop and the external communication, as well as the haptic and graphic rendering, were implemented inside Matlab/Simulink using Handshake proSENSE toolbox, guaranteeing a real-time system. As an example, a five-bar linkage haptic device with two active degrees-of-freedom (DOF) was designed and integrated within the proposed framework, as well as a device for grasping operations. An extension of this standard approach is verified when the therapist is allowed to feel and interact remotely and in real time with the patient. We applied the proposed concept to a single degree-of-freedom master/slave system. One hand orthosis was used as a master device at the therapist's side, while the other was applied to the patient's hand, and used as a slave device. In order to achieve this issue, we proposed two bilateral control systems in order to guarantee an stable interaction between the master and the slave, even in case of variable network conditions (i.e. Internet). By using the master device, the therapist can remotely move the patient's hand and, at the same time, perceive the patient's resistance to the motion, allowing the assessment of important parameters, such as the residual level of spasticity. In this way, it can be remotely assessed the conditions of the patient and consequently can be proposed a proper rehabilitation program.

Hybrid dynamical systems are dynamical systems in which continuous and discrete evolutions coexist and interact. Their twofold nature makes them particularly powerful for both describing and synthesizing complex dynamical behaviors. In this work we exploit this capability for designing innovative control and estimation algorithms that cope with challenges in aerospace applications. In particular: 1. we propose different impulsive control strategies for the problem of close-range rendezvous between two spacecrafts in elliptic orbits; 2. we design a robust time-sub-optimal controller for a class of linear systems emerging in aerospace applications where the control input is limited in magnitude; 3. we synthesize an observer to estimate the speed of rotary systems providing angular measurements that evolve on the unit circle. To this end, we make use of a recent formalism tailored to hybrid dynamical systems for both modeling and proving desirable properties of the proposed algorithms, which are as well confirmed by simulative and experimental validations.

The population of the advanced countries is ageing, with the direct consequence that an increasing number of people will have to live with sensitive, cognitive and physical disabilities. People with impaired physical ability are not confident to move alone, especially in crowded environment and for long journeys, highly reducing the quality of their life. We propose a new generation of robotic walking assistants whose mechanical and electronic components are conceived to optimize the collaboration between the robot and its users. We will apply these general ideas to investigate the interaction between older adults and a robotic walker, named FriWalk, exploiting it either as a navigational or as a rehabilitation aid. For the use of the FriWalk as a navigation assistance, the system guides the user securing high levels of safety, a perfect compliance with the social rules and non-intrusive interaction between human and machine. To this purpose, we developed several guidance systems ranging from completely passive strategies to active solutions exploiting either the rear or the front motors mounted on the robot. The common strategy at the basis of all the algorithms is that the responsibility of the locomotion belongs always to the user, both to increase the mobility of elder users and to enhance their perception of control over the robot. This way the robot intervenes only whenever it is strictly necessary not to mitigate the user safety. Moreover, the robotic walker has been endowed with a tablet and graphical user interface (GUI) which provides the user with the visual indications about the path to follow. Since the FriWalk was developed to suit the needs of users with different deficits, we conducted extensive human-robot interaction (HRI) experiments with elders, complemented with direct interviews of the participants. As concerns the use of the FriWalk as a rehabilitation aid, force sensing to estimate the torques applied by the user and change the user perceived inertia can be exploited by doctors to let the user feel the device heavier or lighter. Moreover, thanks to a new generation of sensors, the device can be exploited in a clinical context to track the performance of the users' rehabilitation exercises, in order to assist nurses and doctors during the hospitalization of older adults.

The introduction of the Brake By Wire is replacing the traditional mechanical control systems with ECUs and it is raising the need to reproduce feelings of eliminated static mechanical components (i.e. hydraulic fluids, pumps and cylinders). Thanks to electromechanical actuators and human-machine interfaces (i.e. active pedal) it is possible to reproduce such feelings and, therefore, arbitrarily change their features. In this way it will be possible to customize the pedal feelings and the vehicle deceleration needed depending on several factors (i.e. surrounding braking scenario, driver characteristics, race vs day-by-day driving condition). Since braking maneuvers are typically critical and involve the driver, the design and development of brake by wire system must start from the consideration of human factors in order to increase acceptance and braking effectiveness. The objective of this research was to redesign the pedal feelings, making them adaptable to the surrounding. Driver acceptance and braking effectiveness could be highly improved by means of adaptive pedal feelings. The starting points of this research were humans factors in the braking domain. Literature and relevant studies have been taken into consideration to put into evidence human mechanisms and behaviors during braking phases. On such basis, two main results have been found out: braking use cases and pedal feeling curves. With regard to the pedal feelings curves, 4 different pedal curves which describe both force on brake pedal travel and acceleration on brake pedal travel are designed. The pedal feeling depends on several factors like the pedal travel, the pedal idle travel, the effort, responsiveness, deceleration perceived, ease of balance (i.e. ease of modulation), gradual braking and so on. Regarding braking use cases, they are described by vehicle data as speed, acceleration, angles and relevant rates, engine rpm, gas and brake pedal position/speed and so on. These use cases have been clustered in order to meet the 4 pedal curves. The research continued on the implementation of a Matlab/Simulink/Stateflow model for the use case recognition. Basing on the vehicle data, the model is able to find out in which use case the vehicle is (parking, low speed maneuvers, emergency, downhill, and so on). Once it finds out the scenario, the model applies the most appropriate pedal feeling curve (both force feedback and deceleration needed). In the end, the model commands an EC brushless motor which is responsible of the changing of static springs force feedback of the pedal. The scenario recognition model has been validated through vehicle data on real road whereas the pedal feeling and relevant motor behaviors have been validate on bench tests.

The following thesis describes the development of a distributed system for the monitoring of numerical control machine tools. This work discusses all the design steps from the conceptualization of the hardware architecture to the development and implementation of monitoring algorithms. In the first chapter the state of art in machining automation will be presented, with a particular focus on monitoring and control system for machining operations. The second chapter describes the architecture of the proposed system both from the hardware and the software side. The third chapter goes into detail of the developed monitoring algorithms intended to the application on milling machines. The implementation and results of experimental tests will be discussed. Finally the fourth chapter proposes a new approach applied to the control of linear motion systems. The work presented in this thesis belongs to the applied research field that aims to enhance the automation level of machine tools by developing innovative techniques for the monitoring and control of machining process. The proposed monitoring system has been developed considering as key requirements the possibility to properly operate in several working conditions, the complete integration with the machine tool structure and the ease of use for unskilled personnel. The developed algorithms include the monitoring and mitigation of cutting vibrations, the detection and diagnosis of faults in spindle bearing and the emergency halt of the machine in case of collision. The resulting system provides a flexible and scalable framework easily adaptable to the specific machine tool and machining application. The monitoring tasks allow a fast setup and their execution is mostly automated, requiring a limited interaction with the machine tool end users. In conclusion the monitoring system improves the automation level of the machine tool providing a better control on the process execution. In addition it facilitates the assessment of the machine behavior allowing the objective evaluation of the operative conditions providing a useful support tool for the machine operator.

In this research work, two ADAS have been proposed, both based on optimal control and manoeuvre jerks as parameters for threat assessment. The first is named “Codriver”, and is a system for driver warning. The second is a sort of completion of the first, since it is designed for autonomous vehicle intervention if the driver does not react to the warnings. The Codriver has been developed by the Mechatronics Group of the University of Trento, which the author is part of, in the framework of the European Project “interactIVe”, to warn the driver for all-around threats safety. It has been then implemented on a real vehicle of Centro Ricerche Fiat, which has been widely tested at the end of the project. On the other hand, for the second system only the main components have been developed by the author during a research period at the University of Tokyo, Japan, and its application is restricted to autonomous obstacle avoidance. In particular, a motion planning algorithm has been used together with a control loop de- signed to execute the planned trajectories. Both systems exploit Optimal Control (OC) for motion planning: the Codriver uses OC to plan real-time ma- noeuvres with humanlike criteria, so that they can be compared to what the driver is doing in order to infer his/her intentions, and warn him if these are not safe; the second system uses OC instead to plan emergency manoeuvres, i.e. neglecting driver actuation limitations and pushing the vehicle towards its physical limits. The initial longitudinal and lateral jerks of the planned manoeuvres are used by both the systems as parameters for risk assessment. Manoeuvre jerks are proportional to pedal and steering wheel velocities, and their initial values thus describe the entity of the correction needed by the driver to achieve a given goal. Since human drivers plan and act with minimum jerk criteria, and are jerk-limited, more and more severe manoeuvres at a given point are not reachable anymore by a human driver, since they require too high initial jerks: initial jerks can be thus considered proportional to the risk level of current situation. For this reason, when the manoeuvres to handle current scenario require jerks beyond a given threshold, the Codriver outputs a warning. This threshold must be lower than driver limits, so that he/she will be able to react to the warning and still have the chance to perform a safe manoeuvre. When the required jerks exceed drivers’ actuation limits, the risk level raises to an upper step, where driver warning would be not effective and autonomous vehicle intervention should be enabled. In obstacle avoidance scenarios, it was demonstrated during driving simulator tests that manoeuvre jerks are more robust parameters for risk assessment than for example time headways, since they are less affected by driver’s age and gender.