publications
publications by categories in reversed chronological order. generated by jekyll-scholar.
Submitted Papers
- Position-based Visual Servo Control without Hand-Eye Calibrationsubmitted to Robotics and Autonomous Systems
We deal with the position-based visual servoing control of a robot manipulator having a camera mounted in an eye-in-hand configuration. The goal is to move the robot in order to reach the desired pose of the camera with respect to the object to be monitored and tracked. We assume that the internal parameters of the camera are known, while the pose of the camera with respect to the robot flange is unknown, except for some rough bounds on the rotation and translation components. By embedding the unknown Jacobian of the flange-to-camera motion in a suitable polytope and exploiting a Lyapunov-based result, we are able to guarantee asymptotic convergence to the desired camera-to-object reference pose. Our experimental results, performed on a 6 degrees-of-freedom (DoF) robotic manipulator, show that the proposed approach is effective in both achieving a single target pose and performing a target object tracking.
Journal Papers
- Model-free kinematic control for robotic systemsAutomatica, 2025
We propose a new model-free approach for kinematic robot control, where both the model and its parameters are partially unknown, which is inspired by the model-free plant tuning framework. The proposed method only relies on the assumption that the relationship between control inputs and outputs is a smooth and static unknown function, whose partial derivatives have lower and upper bounds that are approximately known. Notably, our approach does not require a learning phase, and it is flexible enough to be applied to a wide class of robotic structures. To showcase the methodology, two distinct types of robotic challenges are considered: the control of a family of cable-driven parallel robots and the control of a tendon-driven soft robot. We devise a task-independent approach to synthesize controllers that enable robots to achieve their goals, with minimum prior knowledge of the nonlinear system. Experimental results are presented for the cable-driven parallel robot, while simulations are conducted for the soft robot case.
- Data-driven dynamic relatively optimal controlFelice Andrea Pellegrino, Franco Blanchini, Gianfranco Fenu, and Erica SalvatoEuropean Journal of Control, 20232023 European Control Conference Special Issue
We show how the recent works on data-driven open-loop minimum-energy control for linear systems can be exploited to obtain closed-loop control laws in the form of linear dynamic controllers that are relatively optimal. Besides being stabilizing, they achieve the optimal minimum-energy trajectory when the initial condition is the same as the open-loop optimal control problem. The order of the controller is N−n, where N is the length of the optimal open-loop trajectory, and n is the order of the system. The same idea can be used for obtaining a relatively optimal controller, entirely based on data, from open-loop trajectories starting from up to n linearly independent initial conditions.
- Model-Free Feedback Control Synthesis From Expert DemonstrationFranco Blanchini, Fabrizio Dabbene, Gianfranco Fenu, Felice Andrea Pellegrino, and Erica SalvatoIEEE Control Systems Letters, 2023
We show how it is possible to synthesize a stabilizing feedback control, in the complete absence of a model, starting from the open-loop control generated by an expert operator, capable of driving a system to a spe- cific set-point. We assume that the system is linear and discrete time. We propose two different controls: a linear dynamic and a static, piecewise linear, one. We show the performance of the proposed controllers on a ship steering problem.
- An Artificial Intelligence System for Automatic Recognition of Punches in Fourteenth-Century Panel PaintingMarco Zullich, Vanja Macovaz, Giovanni Pinna, and Felice Andrea PellegrinoIEEE Access, 2023
In Late-Medieval panel paintings from the Tuscan area, mechanical tools called punches were used to impress repeated motifs on gold foils to create decorative patterns. Such patterns can be used as clues to objectively support the attribution of the paintings, as proposed by art historian Erling S. Skaug in his decades-long study on punches. We investigate the feasibility of employing automatic pattern recognition techniques for accelerating the process of classification of punches by experts working in the field. We propose a system composed of (a) a Convolutional Neural Network for categorizing a punch contained in a frame, and (b) an additional component for uncertainty estimation, aimed at recognizing possible Out-of-Distribution (OOD) samples. After collecting a set of 3815 punches from four 14th century panel paintings from Tuscany, we train a Convolutional Neural Network which achieves very high test-set accuracy. As far as the uncertainty estimation is concerned, we experiment with two techniques, OpenGAN and II-loss, both exhibiting very positive results. The former seems to work better on specific data extracted from images of panel paintings, while the latter showcases a more consistent behavior when considering additional OOD data obtained randomly. These outcomes indicate that an application of our system in support of experts is feasible, although we subsequently show that additional experiments on larger datasets might be required.
- Singularity Avoidance for Cart-Mounted Hand-Guided Collaborative Robots: A Variational ApproachErica Salvato, Walter Vanzella, Gianfranco Fenu, and Felice Andrea PellegrinoRobotics, 2022
Most collaborative robots (cobots) can be taught by hand guiding: essentially, by manually jogging the robot, an operator teaches some configurations to be employed as via points. Based on those via points, Cartesian end-effector trajectories such as straight lines, circular arcs or splines are then constructed. Such methods can, in principle, be employed for cart-mounted cobots (i.e., when the jogging involves one or two linear axes, besides the cobot axes). However, in some applications, the sole imposition of via points in Cartesian space is not sufficient. On the contrary, albeit the overall system is redundant, (i)the via points must be reached at the taught joint configurations, and (ii)the undesirable singularity (and near-singularity) conditions must be avoided. The naive approach, consisting of setting the cart trajectory beforehand (for instance, by imposing a linear-in-time motion law that crosses the taught cart configurations), satisfies the first need, but does not guarantee the satisfaction of the second. Here, we propose an approach consisting of (i) a novel strategy for decoupling the planning of the cart trajectory and that of the robot joints, and (ii) a novel variational technique for computing the former in a singularity-aware fashion, ensuring the avoidance of a class of workspace singularity and near-singularity configurations.
- Learning-based automatic classification of lichens from imagesAlberto Presta, Felice Andrea Pellegrino, and Stefano MartellosBiosystems Engineering, 2022
Biomonitoring plays a crucial role in the assessment of air quality, as it allows to estimate the presence of pollutants, by measuring deviations from normality of the components of an ecosystem. Lichens are among the organisms most commonly used as bioindicators. The present study deals with the classification of lichen taxa from images, by means of a machine learning process based on patch classification. A given image is divided in non-overlapping patches, and each of them undergoes feature extraction and classification, eventually being associated to a category. Three different methods for extracting patch descriptors are investigated: (i) handcrafted descriptors based on classical feature extractor algorithms, (ii) convolutional neural networks employed as feature extractors, and (iii) scattering networks, which combine wavelet convolutions and nonlinear operators. For each of these methods, the descriptors are used as inputs for a classification algorithm. The whole process is evaluated in terms of classification accuracy, empirically determining the most appropriate parameters for the different models implemented. By using the dataset of lichens of this study, best results (∼ 0.89 accuracy) are obtained with a specific handcrafted descriptor (dense SIFT), thus providing insights on the kind of representation which is most suitable for the task.
- Merging pruning and neuroevolution: towards robust and efficient controllers for modular soft robotsGiorgia Nadizar, Eric Medvet, Ola Huse Ramstad, Stefano Nichele, Felice Andrea Pellegrino, and Marco ZullichThe Knowledge Engineering Review, 2022
Artificial neural networks (ANNs) can be employed as controllers for robotic agents. Their structure is often com- plex, with many neurons and connections, especially when the robots have many sensors and actuators distributed across their bodies and/or when high expressive power is desirable. Pruning (removing neurons or connections) reduces the complexity of the ANN, thus increasing its energy efficiency, and has been reported to improve the gen- eralization capability, in some cases. In addition, it is well-known that pruning in biological neural networks plays a fundamental role in the development of brains and their ability to learn. In this study, we consider the evolutionary optimization of neural controllers for the case study ofVoxel-based soft robots, a kind of modular, bio-inspired soft robots, applying pruning during fitness evaluation. For a locomotion task, and for centralized as well as distributed controllers, we experimentally characterize the effect of different forms of pruning on after-pruning effectiveness, life-long effectiveness, adaptability to new terrains, and behavior. We find that incorporating some forms of pruning in neuroevolution leads to almost equally effective controllers as those evolved without pruning, with the benefit of higher robustness to pruning. We also observe occasional improvements in generalization ability.
- Machine learning for computationally efficient electrical loads estimation in consumer washing machinesVittorio Casagrande, Gianfranco Fenu, Felice Andrea Pellegrino, Gilberto Pin, Erica Salvato, and Davide ZorzenonNeural Computing and Applications, 2021
Estimating the wear of the single electrical parts of a home appliance without resorting to a large number of sensors is desirable for ensuring a proper level of maintenance by the manufacturers. Deep learning techniques can be effective tools for such estimation from relatively poor measurements, but their computational demands must be carefully considered, for the actual deployment. In this work, we employ one-dimensional Convolutional Neural Networks and Long Short-Term Memory networks to infer the status of some electrical components of different models of washing machines, from the electrical signals measured at the plug. These tools are trained and tested on a large dataset (502 washing cycles) collected from four different washing machines and are carefully designed in order to comply with the memory constraints imposed by available hardware selected for a real implementation. The approach is end-to-end; i.e., it does not require any feature extraction, except the harmonic decomposition of the electrical signals, and thus it can be easily generalized to other appliances.
- Crossing the Reality Gap: a Survey on Sim-to-Real Transferability of Robot Controllers in Reinforcement LearningErica Salvato, Gianfranco Fenu, Eric Medvet, and Felice Andrea PellegrinoIEEE Access, 2021
The growing demand for robots able to act autonomously in complex scenarios has widely accelerated the introduction of Reinforcement Learning (RL) in robots control applications. However, the trial and error intrinsic nature of RL may result in long training time on real robots and, moreover, it may lead to dangerous outcomes. While simulators are useful tools to accelerate RL training and to ensure safety, they often are provided only with an approximated model of robot dynamics and of its interaction with the surrounding environment, thus resulting in what is called the reality gap (RG): a mismatch of simulated and real control-law performances caused by the inaccurate representation of the real environment in simulation. The most undesirable result occurs when the controller learnt in simulation fails the task on the real robot, thus resulting in an unsuccessful sim-to-real transfer. The goal of the present survey is threefold: (1) to identify the main approaches to face the RG problem in the context of robot control with RL, (2) to point out their shortcomings, and (3) to outline new potential research areas.
- An Online Iterative Linear Quadratic Approach for a Satisfactory Working Point Attainment at FERMINiky Bruchon, Gianfranco Fenu, Giulio Gaio, Simon Hirlander, Marco Lonza, Felice Andrea Pellegrino, and Erica SalvatoInformation, 2021
The attainment of a satisfactory operating point is one of the main problems in the tuning of particle accelerators. These are extremely complex facilities, characterized by the absence of a model that accurately describes their dynamics, and by an often persistent noise which, along with machine drifts, affects their behaviour in unpredictable ways. In this paper, we propose an online iterative Linear Quadratic Regulator (iLQR) approach to tackle this problem on the FERMI free-electron laser of Elettra Sincrotrone Trieste. It consists of a model identification performed by a neural network trained on data collected from the real facility, followed by the application of the iLQR in a Model-Predictive Control fashion. We perform several experiments, training the neural network with increasing amount of data, in order to understand what level of model accuracy is needed to accomplish the task. We empirically show that the online iLQR results, on average, in fewer steps than a simple gradient ascent (GA), and requires a less accurate neural network to achieve the goal.
- Model Predictive Control of Glucose Concentration Based on Signal Temporal Logic Specifications with Unknown-Meals OccurrenceFrancesca Cairoli, Gianfranco Fenu, Felice Andrea Pellegrino, and Erica SalvatoCybernetics and Systems, 2020
The glycemia regulation is a significant challenge in the Artificial Pancreas (AP) scenario. Several control systems have been developed in the last years, many of them requiring meal announcements. Therefore, if the patients skip the meal announcement or make a mistake in the estimation of the amount of carbohydrates, the control performance will be negatively affected. In this extended version of our previous work, we present a Model Predictive Controller (MPC) for the AP in which the meal is treated as a disturbance to be estimated by an Unknown Input Observer (UIO). The MPC constraints are expressed in terms of Signal Temporal Logic (STL) specifications. Indeed, in the AP some requirements result in hard constraints (in particular, absolutely avoid hypoglycemia and absolutely avoid severe hyperglycemia) and some other in soft constraints (avoid a prolonged hyperglycemia) and STL is suitable for expressing such requirements. The achieved results are obtained using the BluSTL toolbox, which allows to synthesize model predictive controllers with STL constraints. We report simulations showing that the proposed approach, avoiding unnecessary restrictions, provides safe trajectories in correspondence of higher unknown disturbance.
- Basic Reinforcement Learning Techniques to Control the Intensity of a Seeded Free-Electron LaserNiky Bruchon, Gianfranco Fenu, Giulio Gaio, Marco Lonza, Finn Henry O’Shea, Felice Andrea Pellegrino, and Erica SalvatoElectronics, 2020
Optimal tuning of particle accelerators is a challenging task. Many different approaches have been proposed in the past to solve two main problems—attainment of an optimal working point and performance recovery after machine drifts. The most classical model-free techniques (e.g., Gradient Ascent or Extremum Seeking algorithms) have some intrinsic limitations. To overcome those limitations, Machine Learning tools, in particular Reinforcement Learning (RL), are attracting more and more attention in the particle accelerator community. We investigate the feasibility of RL model-free approaches to align the seed laser, as well as other service lasers, at FERMI, the free-electron laser facility at Elettra Sincrotrone Trieste. We apply two different techniques—the first, based on the episodic Q-learning with linear function approximation, for performance optimization; the second, based on the continuous Natural Policy Gradient REINFORCE algorithm, for performance recovery. Despite the simplicity of these approaches, we report satisfactory preliminary results, that represent the first step toward a new fully automatic procedure for the alignment of the seed laser to the electron beam. Such an alignment is, at present, performed manually.
- Support Vector Representation Machine for superalloy investment casting optimizationCarmen Del Vecchio, Gianfranco Fenu, Felice Andrea Pellegrino, Michele Di Foggia, Massimo Quatrale, Luca Benincasa, Stefania Iannuzzi, Alessandro Acernese, Pasquale Correra, and Luigi GlielmoApplied Mathematical Modelling, 2019
Machine learning techniques have been widely applied to production processes with the aim of improving product quality, supporting decision-making, or implementing process diagnostics. These techniques proved particularly useful in the investment casting manufacturing industry, where huge variety of heterogeneous data, related to different production processes, can be gathered and recorded but where traditional models fail due to the complexity of the production process. In this study, we apply Support Vector Representation Machine to production data from a manufacturing plant producing turbine blades through investment casting. We obtain an instance ranking that may be used to infer proper values of process parameter set-points.
- A control system for preventing cavitation of centrifugal pumpsValentino Cucit, Fabio Burlon, Gianfranco Fenu, Riccardo Furlanetto, Valentino Cucit, Felice Andrea Pellegrino, and Michele SimonatoEnergy Procedia, 2018
Cavitation is a well-known phenomenon that may occur, among other turbo-machines, in centrifugal pumps and can result in severe damage of both the pump and the whole hydraulic system. There are situations in which, in principle, the cavitation could be avoided by detecting the condition of incipient cavitation, and changing slightly the working point of the whole system in order to move away from that condition. In the present paper two simple closed-loop control strategies are implemented, acting on the pump’s rotational speed and fed by the measurements of a set of inertial sensors. In particular, the research is focused on a centrifugal pump normally employed in hydraulic systems. The pump operates in a dedicated test rig, where cavitation can be induced by acting on a reservoir’s pressure. Three accelerometers are installed in the pump body along three orthogonal axes. An extensive set of experiments has been carried out at different flow rates and a number of signals’ features both in the time domain and in the frequency domain have been considered as indicators of incipient cavitation. The amount of energy of the signal captured by the accelerometer in the component orthogonal to the flow direction, in the band from 10 to 12.8 kHz, demonstrated to be effective in detecting the incipient cavitation, by selecting a proper (condition-dependent) threshold. Therefore, two simple controllers have been designed: the first regulates the speed of the pump, to recover from cavitation, bringing the indicator back to the nominal value, while the second allows to reduce the pump’s rotational speed when the cavitation detector indicates the incipient cavitation and restoring the nominal speed when possible. The latter approach is rather general, because the threshold-based detector can be substituted by any detector providing binary output. Experimental results are reported that demonstrate the effectiveness of the approach.
- Quality of images with toric intraocular lensesDaniele Tognetto, Alberto Armando Perrotta, Francesco Bauci, Silvia Rinaldi, Manlio Antonuccio, Felice Andrea Pellegrino, Gianfranco Fenu, George Stamatelatos, and Noel AlpinsJournal of Cartaract & Refractive Surgery, 2018
- Free-electron laser spectrum evaluation and automatic optimizationNiky Bruchon, Gianfranco Fenu, Giulio Gaio, Marco Lonza, Felice Andrea Pellegrino, and Lorenzo SauleNuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, 2017
The radiation generated by a seeded free-electron laser (FEL) is characterized by a high temporal coherence, which is close to the Fourier limit in the ideal case. The setup and optimization of a FEL is a non-trivial and challenging operation. This is due to the plethora of highly sensitive machine parameters and to the complex correlations between them. The fine tuning of the FEL process is normally supervised by physicists and is carried out by scanning various parameters with the aim of optimizing the spectrum of the emitted pulses in terms of intensity and line-width. In this article we introduce a novel quantitative method for the evaluation of the FEL spectrum via a quality index. Moreover, we investigate the possibility of optimization of the FEL parameters using this index as the objective function of an automatic procedure. We also present the results of the preliminary tests performed in the FERMI FEL focused on the effectiveness and ability of the automatic procedure to assist in the task of machine tuning and optimization.
- Hamiltonian path planning in constrained workspaceDaniele Casagrande, Gianfranco Fenu, and Felice Andrea PellegrinoEuropean Journal of Control, 2017
Abstract A methodology to plan the trajectories of robots that move in an n-dimensional Euclidean space, have to reach a target avoiding obstacles and are constrained to move in a region of the space is described. It is shown that if the positions of the obstacles are known then a Hamiltonian function can be constructed and used to define a collision-free trajectory. It is also shown that the method can be extended to the case in which the target or the obstacles (or both) move. Results of simulations for a pair of planar robots and a three degrees-of-freedom manipulator are finally reported.
- A convex programming approach to the inverse kinematics problem for manipulators under constraintsFranco Blanchini, Gianfranco Fenu, Giulia Giordano, and Felice Andrea PellegrinoEuropean Journal of Control, 2017
Abstract We propose a novel approach to the problem of inverse kinematics for possibly redundant planar manipulators. We show that, by considering the joints as point masses in a fictitious gravity field, and by adding proper constraints to take into account the length of the links, the kinematic inversion may be cast as a convex programming problem. Convex constraints in the decision variables (in particular, linear constraints in the workspace) are easily managed with the proposed approach. We also show how to exploit the idea for avoiding obstacles while tracking a reference end-effector trajectory and discuss how to extend the results to some kinds of non-planar manipulators. Simulation results are reported, showing the effectiveness of the approach.
- Model-Free Plant TuningFranco Blanchini, Gianfranco Fenu, Giulia Giordano, and Felice Andrea PellegrinoIEEE Transactions on Automatic Control, 2017
Given a static plant described by a differentiable input-output function, which is completely unknown, but whose Jacobian takes values in a known polytope in the matrix space, this paper considers the problem of tuning (i.e., driving to a desired value) the output, by suitably choosing the input. It is shown that, if the polytope is robustly nonsingular (or has full rank, in the nonsquare case), then a suitable tuning scheme drives the output to the desired point. The proof exploits a Lyapunov-like function and applies a well-known game-theoretic result, concerning the existence of a saddle point for a min-max zero-sum game. When the plant output is represented in an implicit form, it is shown that the same result can be obtained, resorting to a different Lyapunov-like function. The case in which proper input or output constraints must be enforced during the transient is considered as well. Some application examples are proposed to show the effectiveness of the approach.
- A Youla–Kučera parameterization approach to output feedback relatively optimal controlFranco Blanchini, Patrizio Colaneri, Yasumasa Fujisaki, Stefano Miani, and Felice Andrea PellegrinoSystems & Control Letters, 2015
This paper presents a continuous time solution to the problem of designing a relatively optimal control, precisely, a dynamic control which is optimal with respect to a given initial condition and is stabilizing for any other initial state. This technique provides a drastic reduction of the complexity of the controller and successfully applies to systems in which (constrained) optimality is necessary for some “nominal operation” only. The technique is combined with a pole assignment procedure. It is shown that once the closed-loop poles have been fixed and an optimal trajectory originating from the nominal initial state compatible with these poles is computed, a stabilizing compensator which drives the system along this trajectory can be derived in closed form. There is no restriction on the optimality criterion and the constraints. The optimization is carried out over a finite-dimensional parameterization of the trajectories. The technique has been presented for state feedback. We propose here a technique based on the Youla–Kučera parameterization which works for output feedback. The main result is that we provide conditions for solvability in terms of a set of linear algebraic equations.
- Approximate model predictive control laws for constrained nonlinear discrete–time systems: analysis and offline designGilberto Pin, Marco Filippo, Felice Andrea Pellegrino, Gianfranco Fenu, and Thomas ParisiniInternational Journal of Control, 2013
The objective of this work consists in the offline approximation of possibly discontinuous model predictive control laws for nonlinear discrete-time systems, while enforcing hard constraints on state and input variables. Obtaining an offline approximation of the receding horizon control law may lead to a very significant reduction of the online computational burden with respect to algorithms based on iterated optimization, thus allowing the application to fast dynamics plants. The proposed approximation scheme allows to cope with discontinuous control laws, such as those arising from constrained nonlinear finite horizon optimal control problems. A detailed stability analysis of the closed-loop system driven by the approximated state-feedback controller shows that the devised technique guarantees the input-to-state practical stability with respect to the (non-fading) approximation-induced errors. Two examples are provided to show the effectiveness of the method when the approximator is chosen either as a discontinuous nearest point function or as a smooth neural network.
- Disturbance-driven model predictive control by means of Youla-Kučera parameter switching with an application to drainage canal controlFranco Blanchini, Felice Andrea Pellegrino, and Stefano MianiInternational Journal of Robust and Nonlinear Control, 2012
Motivated by applications such as drainage canal control, where the disturbance acting on a system can be predicted on the basis of weather forecast, we propose a model predictive control technique consisting in the online tuning of a Youla–Kucera parameter. Essentially, the optimal control sequence is supplied to the system by properly setting the coefficients of the Youla–Kucera parameter taken as a finite impulse response, whose length depends on that of the optimal trajectory. Stability is guaranteed for any arbitrary choice of the time-varying coefficients that are adapted to the current forecast of disturbance. The approach is illustrated by numerical simulations of a drainage canal control system.
- High-Gain Adaptive Control: A Derivative-Based ApproachFranco Blanchini, Thomas Parisini, Felice Andrea Pellegrino, and Gilberto PinIEEE Transactions on Automatic Control, Sep 2009
In this work, we propose an adaptive scheme which is a counterpart of existing high gain control techniques based on control Lyapunov functions. Given a control Lyapunov function, the main idea is that of tuning the feedback gain according to a suitably-chosen Lyapunov time-derivative. The control gain is not monotonically non-decreasing as in existing techniques, but it is increased or decreased depending on the imposed derivative, thus avoiding the well-known issue of actuator over-exploitation. We are able to show robust convergence of the proposed adaptive control scheme as well as other interesting properties. For instance, it is possible to guarantee an a-priori given upper bound for the transient mode of behavior during adaptation. Furthermore, if the control Lyapunov function is designed based on an optimal control problem, then the control action is nominally optimal, precisely it yields the optimal trajectory for any initial condition, if the actual plant matches the nominal system.
- Enhancing Controller Performance for Robot Positioning in a Constrained EnvironmentFranco Blanchini, Stefano Miani, Felice Andrea Pellegrino, and Bart Van ArkelIEEE Transactions on Control Systems Technology, 2008
The paper deals with the problem of positioning a manipulator in a cluttered environment while avoiding collision with obstacles. Recently a control strategy based on invariant sets has been introduced by some of the authors: it consists of covering the configuration space by means of a connected family of poly- hedral regions which can be rendered controlled-invariant. Each of these regions includes some crossing points to the confining (and partially overlapping) regions. The control is hierarchically structured: a high-level controller establishes a proper sequence of regions to be crossed to reach the one in which the target con- figuration is included. A low-level controller solves the problem of tracking, within a region, the crossing point to the next con- fining region and, eventually, tracking the reference whenever it is included in the current one. Here we focus on the low-level controller, providing two novel contributions: first we extend the previous results, based on a vertex representation of the polyhedral sets, to the face representation which is more natural and offers significant computational advantages for on-line implementation; second, we provide a new low-level speed-saturated controller in order to improve the performance of the previous one in terms of convergence speed. We also investigate the robustness of the proposed controller. Experimental results on a Cartesian robot are provided.
- Simultaneous performance achievement via compensator blendingFranco Blanchini, Patrizio Colaneri, and Felice Andrea PellegrinoAutomatica, Jan 2008
In this paper we consider the problem of designing a state-feedback controller that simultaneously achieves different optimality criteria defined on different input–output pairs. Precisely, if r “optimal” target transfer functions are given (as the result of local “optimal” controllers), it is shown that (under mild assumptions) there exists a unique controller capable of replicating these transfer functions in the closed-loop system, so simultaneously achieving the performances inherited by the chosen local transfer functions. An explicit and constructive procedure (we refer to such procedure as “compensator blending”) is provided. The possibility of designing a stable blending compensator or the generalization to dynamic local controllers or time varying systems are also discussed. We finally consider the dual version of the problem, precisely, we show how to achieve simultaneous optimality by filter blending.
- Relatively Optimal Control: A Static Piecewise-Affine SolutionFranco Blanchini, and Felice Andrea PellegrinoSIAM Journal on Control and Optimization, 2007
A relatively optimal control is a stabilizing controller that, without initialization nor feedforwarding and tracking the optimal trajectory, produces the optimal (constrained) behavior for the nominal initial condition of the plant. In a previous work, for discrete-time linear systems, we presented a linear dynamic relatively optimal control. Here we provide a static solution, namely a deadbeat piecewise-affine state-feedback controller based on a suitable partition of the state space into polyhedral sets. The vertices of the polyhedra are the states of the optimal trajectory; hence a bound for the complexity of the controller is known in advance. We also show how to obtain a controller that is not deadbeat by removing the zero terminal constraint while guaranteeing stability. Finally, we compare the proposed static compensator with the dynamic one.
- HAB Buoy: a new instrument for in situ monitoring and early warning of harmful algal bloom eventsP.F. Culverhouse, R. Williams, B. Simpson, C. Gallienne, B. Reguera, M. Cabrini, S. Fonda-Umani, Thomas Parisini, Felice Andrea Pellegrino, Y. Pazos, H. Wang, L. Escalera, A. Moroño, M. Hensey, J. Silke, and 6 more authorsAfrican Journal of Marine Science, 2006
A new microplankton imaging and analysis instrument, HAB Buoy, is described. It integrates a high-speed camera for in-flow image acquisition with automatic specimen labelling software, known as DiCANN (Dino- flagellate Categorisation by Artificial Neural Network). Some preliminary results are presented together with a rationale for its use.
- Relatively Optimal Control With Characteristic Polynomial Assignment and Output FeedbackFranco Blanchini, and Felice Andrea PellegrinoIEEE Transactions on Automatic Control, 2006
A relatively optimal control is a stabilizing controller such that, if initialized at its zero state, produces the optimal (con- strained) behavior for the nominal initial condition of the plant (without feedforwarding and tracking the optimal trajectory). In this paper, we prove that a relatively optimal control can be ob- tained under quite general constraints and objective function, in particular without imposing 0-terminal constraints as previously done. The main result is that stability of the closed-loop system can be achieved by assigning an arbitrary closed-loop character- istic stable polynomial to the plant. An explicit solution is provided. We also show how to choose the characteristic polynomial in such a way that the constraints (which are enforced on a finite horizon) can be globally or ultimately satisfied (i.e., satisfied from a certain time on). We provide conditions to achieve strong stabilization (sta- bilization by means of a stable compensator) precisely, we show how to assign both compensator and closed-loop poles. We consider the output feedback problem, and we show that it can be success- fully solved by means of a proper observer initialization (based on output measurements only). We discuss several applications of the technique and provide experimental results on a cart-pendulum system.
- Self-Adaptive RegularizationWalter Vanzella, Felice Andrea Pellegrino, and Vincent TorreIEEE Transactions on Pattern Analysis and Machine Intelligence, 2004
Often an image g(x,y) is regularized and even restored by minimizing the Mumford-Shah functional. Properties of the regularized image u(x,y) depends critically on the numerical value of the two parameters alpha and gamma controlling smoothness and fidelity. When alpha and gamma are constant over the image, small details are lost when an extensive filtering is used in order to remove noise. In this paper, it is shown how the two parameters alpha and gamma can be made self-adaptive. In fact, alpha and gamma are not constant but automatically adapt to the local scale and contrast of features in the image. In this way, edges at all scales are detected and boundaries are well-localized and preserved. In order to preserve trihedral junctions alpha and gamma become locally small and the regularized image u(x,y) maintains sharp and well-defined trihedral junctions. Images regularized by the proposed procedure are well-suited for further processing, such as image segmentation and object recognition.
- Automatic Visual Recognition of Deformable Objects for Grasping and ManipulationGian Luca Foresti, and Felice Andrea PellegrinoIEEE Transactions on Systems, Man and Cybernetics, Part C (Applications and Reviews), 2004
This paper describes a vision-based system that is able to automatically recognize deformable objects, to estimate their pose, and to select suitable picking points. A hierarchical self-organized neural network is used to segment color images based on texture information. A morphological analysis allows the recognition of the objects and the picking points extraction. The proposed approach is useful in all of the situations where texture properties are significant for detecting regions of interest on deformable objects. Several tests on a large number of images, acquired in real operative working conditions, demonstrate the effectiveness of the system.
- Edge detection revisitedFelice Andrea Pellegrino, Walter Vanzella, and Vincent TorreIEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), 2004
The present manuscript aims at solving four problems of edge detection: the simultaneous detection of all step edges from a fine to a coarse scale; the detection of thin bars with a width of very few pixels; the detection of trihedral junctions; the development of an algorithm with image-independent parameters. The proposed solution of these problems combines an extensive spatial filtering with classical methods of computer vision and newly developed algorithms. Step edges are computed by extracting local maxima from the energy summed over a large bank of directional odd filters with a different scale. Thin roof edges are computed by considering maxima of the energy summed over narrow odd and even filters along the direction providing maximal response. Junctions are precisely detected and recovered using the output of directional filters. The proposed algorithm has a threshold for the minimum contrast of detected edges: for the large number of tested images this threshold was fixed equal to three times the standard deviation of the noise present in usual acquisition system (estimated to be between 1 and 1.3 gray levels out of 256), therefore, the proposed scheme is in fact parameter free. This scheme for edge detection performs better than the classical Canny edge detector in two quantitative comparisons: the recovery of the original image from the edge map and the structure from motion task. As the Canny detector in previous comparisons was shown to be the best or among the best detectors, the proposed scheme represents a significant improvement over previous approaches.
- Control of manipulators in a constrained workspace by means of linked invariant setsFranco Blanchini, Felice Andrea Pellegrino, and Luca VisentiniInternational Journal of Robust and Nonlinear Control, 2004
We propose a new technique for controlling manipulators in constrained environments. Based on recent developments on constrained control theory, our approach basically consists in covering the admissible region of the configuration space by partially overlapping convex polyhedra arbitrarily fixed and forming a connected family. Each of these polyhedra, defined in the configuration space, is suitably extended in the state-space (i.e. configuration-plus-velocity space) in order to maintain all the original connections among regions and to be a tracking domain of attraction for the system, i.e. a set of initial states from which a reference signal can be asymptotically approached without constraints violation during the transient. The connection path is not generated a priori, but it is automatically produced on-line by a hierarchical feedback controller. A high-level controller selects the confining polyhedron, to which the current state has to be transferred, which is the closest to the one containing the target reference. A low-level controller solves the problem of locally tracking a suitable crossing point to the desired new region under constraints. The robustness of the scheme as well as the control effort constraints are also taken into account.
- Suboptimal Receding Horizon Control for Continuous-Time SystemsFranco Blanchini, Stefano Miani, and Felice Andrea PellegrinoIEEE Transactions on Automatic Control, 2003
Solving a continuous-time optimal control problem under state and control constraints is known to be a very hard task. In this note, we propose a suboptimal solution based on the Euler auxiliary system (EAS). We show that we can determine a continuous-time stabilizing control whose cost not only converges to the optimal as the EAS time parameter vanishes, but it is also upper bounded by the discrete-time cost, no matter how such a parameter is chosen. In particular, continuous-time linear problems with convex cost can be solved by considering a fictitious receding-horizon scheme. Both stability and constraints satisfaction are guaranteed for the continuous-time system. This scheme turns out to be very useful when, due to unstable or poorly damped dynamics, the digital implementation of the control requires a very small (virtually zero) sampling time, since the “time parameter” of the EAS can be much greater than the sampling time, without compromising stability, with a strongly reduced computational burden.
- Relatively Optimal Control and Its Linear ImplementationFranco Blanchini, and Felice Andrea PellegrinoIEEE Transactions on Automatic Control, 2003
Motivated by the fact that determining a feedback solution for the optimal control problem under constraints is a hard task we introduce the concept of relative optimality, roughly optimality for a specific (nominal) plant initial condition. We consider a generic discrete-time finite-horizon constrained optimal control problem for linear systems, and we seek for a state feedback (possibly dynamic) controller. As a fundamental requirement, we do not admit preactions or controller-state initialization based on the plant initial state and we assume our controller to be time-invariant. In particular, we do not consider controllers simply achieved by the feedforward and tracking of the optimal trajectory. A relatively optimal control is a stabilizing controller such that, if initialized at its zero state, produces the optimal (constrained) trajectory for the nominal initial condition of the plant. We show that one of such controllers is linear, dead-beat, and its order is equal to the length of the horizon minus the plant order, thus, of complexity which is known a priori. Some additional features such as the assignment of the compensator poles to achieve strong stabilization are proposed. We show that, by means of the proposed approach, we can face several problems such as optimal point-to-point operations, optimal impulse response and optimal tracking.
Conference Papers
- An Active Disturbance Rejection Model Predictive Controller for Constrained Over-Actuated SystemsErica Salvato, Gianfranco Fenu, Felice Andrea Pellegrino, and Thomas ParisiniIn 2024 European Control Conference (ECC), 2024
- Model predictive control for temperature regulation of professional ovensJuan Marcelo Castellino, Francesco Forte, Gianfranco Fenu, and Felice Andrea PellegrinoIn 2023 9th International Conference on Control, Decision and Information Technologies (CoDIT), Jul 2023
We apply the model predictive control (MPC) strategy in an industrial setting, specifically for controlling the temperature of Combi Oven Professional Appliances. The proposed method takes into account input and output constraints, as well as the presence of multiple sources of disturbance. The workflow includes identifying and validating a model of the cell temperature and incorporating disturbance models. MPC is implemented using a state-space formulation. The proposed method shows significant energy saving and tracking error reduction with respect to the current oven control; its effectiveness has been demonstrated through several tests carried out on a professional oven.
- Explainable Automated Anomaly Recognition in Failure Analysis: is Deep Learning Doing it Correctly?Leonardo Arrighi, Sylvio Barbon Junior, Felice Andrea Pellegrino, Michele Simonato, and Marco ZullichIn Explainable Artificial Intelligence, 2023
EXplainable AI (XAI) techniques can be employed to help identify points of concern in the objects analyzed when using image-based Deep Neural Networks (DNNs). There has been an increasing number of works proposing the usage of DNNs to perform Failure Analysis (FA) in various industrial applications. These DNNs support practitioners by providing an initial screening to speed up the manual FA process. In this work, we offer a proof-of-concept for using a DNN to recognize failures in pictures of Printed Circuit Boards (PCBs), using the boolean information of (non) faultiness as ground truth. To understand if the model correctly identifies faulty connectors within the PCBs, we make use of XAI tools based on Class Activation Mapping (CAM), observing that the output of these techniques seems not to align well with these connectors. We further analyze the faithfulness of these techniques with respect to the DNN, observing that often they do not seem to capture relevant features according to the model’s decision process. Finally, we mask out faulty connectors from the original images, noticing that the DNN predictions do not change significantly, thus showing that the model possibly did not learn to base its predictions on features associated with actual failures. We conclude with a warning that FA using DNNs should be conducted using more complex techniques, such as object detection, and that XAI tools should not be taken as oracles, but their correctness should be further analyzed.
- Closed-loop Control from Data-Driven Open-Loop Optimal Control TrajectoriesFelice Andrea Pellegrino, Franco Blanchini, Gianfranco Fenu, and Erica SalvatoIn 2022 European Control Conference (ECC), Jul 2022
We show how the recent works on data driven open-loop minimum-energy control for linear systems can be exploited to obtain closed-loop piecewise-affine control laws, by employing a state-space partitioning technique which is at the basis of the static relatively optimal control. In addition, we propose a way for employing portions of the experimental input and state trajectories to recover information about the natural movement of the state and dealing with non-zero initial conditions. The same idea can be used for formulating several open-loop control problems entirely based on data, possibly including input and state constraints.
- YOLO-Based Face Mask Detection on Low-End Devices Using Pruning and QuantizationBenedetta Liberatori, Ciro Antonio Mami, Giovanni Santacatterina, Marco Zullich, and Felice Andrea PellegrinoIn 2022 45th International Convention on Information, Communication and Electronic Technology (MIPRO), May 2022
Deploying Deep Learning (DL) based object detection (OD) models in low-end devices, such as single board computers, may lead to poor performance in terms of frames-per-second (FPS). Pruning and quantization are well-known compression techniques that can potentially lead to a reduction of the computational burden of a DL model, with a possible decrease of performance in terms of detection accuracy. Motivated by the widespread introduction of face mask mandates by many institutions during the Covid-19 pandemic, we aim at training and compressing an OD model based on YOLOv4 to recognize the presence of face masks, to be deployed on a Raspberry Pi 4. We investigate the capability of different kinds of pruning and quantization techniques of increasing the FPS with respect to the uncompressed model, while retaining the detection accuracy. We quantitatively assess the pruned and quantized models in terms of Mean Average Precision (mAP) and FPS, and show that with proper pruning and quantization, the FPS can be doubled with a moderate loss in mAP. The results provide guidelines for compression of other OD models based on YOLO.
- Characterization of Modeling Errors Affecting Performances of a Robotics Deep Reinforcement Learning Controller in a Sim-to-Real TransferErica Salvato, Gianfranco Fenu, Eric Medvet, and Felice Andrea PellegrinoIn 2021 44th International Convention on Information, Communication and Electronic Technology (MIPRO), 2021
Simulation is a powerful tool used to train Re- inforcement Learning (RL) agents involved in robotic tasks. It allows to collect large amount of data in comparatively faster and safer way than on the real robot. However, a simulator is only an approximation of the physical system to be controlled. Due to modeling errors, a controller learned on the simulator dynamics may behave differently once applied to the real robot. In the worst case, the controller, although being successful when applied on the simulator, may fail when applied on the real platform. In this paper, we deal with the sim-to-real transfer of a RL controller for a Poppy Ergo-Jr robotic arm involved in a positioning task: i.e., moving the servo joints in order to reach a desired target position with the end-effector. In particular, we want to investigate the differences between the real robot and its simulator and how they affect the controller performance after its transfer from the simulator to the real platform.
- Speeding-up pruning for Artificial Neural Networks: Introducing Accelerated Iterative Magnitude PruningMarco Zullich, Eric Medvet, Felice Andrea Pellegrino, and Alessio AnsuiniIn 2020 25th International Conference on Pattern Recognition (ICPR), Jan 2021
In recent years, Artificial Neural Networks (ANNs) pruning has become the focal point of many researches, due to the extreme overparametrization of such models. This has urged the scientific world to investigate methods for the simplification of the structure of weights in ANNs, mainly in an effort to reduce time for both training and inference. Frankle and Carbin [1], and later Renda, Frankle, and Carbin [2] introduced and refined an iterative pruning method which is able to effectively prune the network of a great portion of its parameters with little to no loss in performance. On the downside, this method requires a large amount of time for its application, since, for each iteration, the network has to be trained for (almost) the same amount of epochs of the unpruned network. In this work, we show that, for a limited setting, if targeting high overall sparsity rates, this time can be effectively reduced for each iteration, save for the last one, by more than 50%, while yielding a final product (i.e., final pruned network) whose performance is comparable to the ANN obtained using the existing method.
- On the Effects of Pruning on Evolved Neural Controllers for Soft RobotsIn Proceedings of the Genetic and Evolutionary Computation Conference Companion, 2021
Artificial neural networks (ANNs) are commonly used for controlling robotic agents. For robots with many sensors and actuators, ANNs can be very complex, with many neurons and connections. Removal of neurons or connections, i.e., pruning, may be desirable because (a) it reduces the complexity of the ANN, making its operation more energy efficient, and (b) it might improve the generalization ability of the ANN. Whether these goals can actually be achieved in practice is however still not well known. On the other hand, it is widely recognized that pruning in biological neural networks plays a fundamental role in the development of brains and their ability to learn. In this work, we consider the case of Voxel-based Soft Robots, a kind of robots where sensors and actuators are distributed over the body and that can be controlled with ANNs optimized by means of neuroevolution. We experimentally characterize the effect of different forms of pruning on the effectiveness of neuroevolution, also in terms of generalization ability of the evolved ANNs. We find that, with some forms of pruning, a large portion of the connections can be pruned without strongly affecting robot capabilities. We also observe sporadic improvements in generalization ability.
- Virtual Redundancy and Barrier Functions for Collision Avoidance in Robotic ManufacturingFelice Andrea Pellegrino, and Walter VanzellaIn 2020 7th International Conference on Control, Decision and Information Technologies (CoDIT), Jun 2020
In some robotic manufacturing processes, a displacement between the nominal tool trajectory and the actual can be tolerated, provided that the displacement occurs along a specific direction/rotation axis. For instance, in wire sawing, a translation along the wire axis is permitted. Such an admissible displacement can be thought of as an additional degree of freedom that could be exploited for accomplishing further tasks, e.g., collision avoidance. In the case of offline programming, the operator can manually adapt the trajectory in order to exploit such additional degree of freedom. However, when the tool trajectory is automatically generated, as in modern flexible automation systems, an automatic way to adjust the displacement is needed. We propose an approach that, based on virtual redundancy (introduced originally for singularity avoidance purposes) and barrier functions, can automatically exploit the additional degree of freedom while satisfying the constraints on the admissible displacement.
- Loads Estimation using Deep Learning Techniques in Consumer Washing MachinesAlexander Babichev, Vittorio Casagrande, Luca Della Schiava, Gianfranco Fenu, Imola Fodor, Enrico Marson, Felice Andrea Pellegrino, Gilberto Pin, Erica Salvato, Michele Toppano, and Davide ZorzenonIn Proceedings of the 9th International Conference on Pattern Recognition Applications and Methods, Feb 2020
Home appliances are nowadays present in every house. In order to ensure a suitable level of maintenance, manufacturers strive to design a method to estimate the wear of the single electrical parts composing an appliance without providing it with a large number of expensive sensors. With this in mind, our goal consists in inferring the status of the electrical actuators of a washing machine, given the measures of electrical signals at the plug, which carry an aggregate information. The approach is end-to-end, i.e. it does not require any feature extraction and thus it can be easily generalized to other appliances. Two different techniques have been investigated: Convolutional Neural Networks and Long Short-Term Memories. These tools are trained and tested on data collected on four different washing machines.
- Mosaic Images Segmentation using U-netGianfranco Fenu, Eric Medvet, Daniele Panfilo, and Felice Andrea PellegrinoIn Proceedings of the 9th International Conference on Pattern Recognition Applications and Methods (ICPRAM 2020), Feb 2020
We consider the task of segmentation of images of mosaics, where the goal is to segment the image in such a way that each region corresponds exactly to one tile of the mosaic. We propose to use a recent deep learning technique based on a kind of convolutional neural networks, called U-net, that proved to be effective in seg- mentation tasks. Our method includes a preprocessing phase that allows to learn a U-net despite the scarcity of labeled data, which reflects the peculiarity of the task, in which manual annotation is, in general, costly. We experimentally evaluate our method and compare it against the few other methods for mosaic images segmen- tation using a set of performance indexes, previously proposed for this task, computed using 11 images of real mosaics. In our results, U-net compares favorably with previous methods. Interestingly, the considered meth- ods make errors of different kinds, consistently with the fact that they are based on different assumptions and techniques. This finding suggests that combining different approaches might lead to an even more effective segmentation.
- On the Similarity between Hidden Layers of Pruned and Unpruned Convolutional Neural NetworksAlessio Ansuini, Eric Medvet, Felice Andrea Pellegrino, and Marco ZullichIn Proceedings of the 9th International Conference on Pattern Recognition Applications and Methods (ICPRAM 2020), Feb 2020
During the last few decades, artificial neural networks (ANN) have achieved an enormous success in regres- sion and classification tasks. The empirical success has not been matched with an equally strong theoretical understanding of such models, as some of their working principles (training dynamics, generalization proper- ties, and the structure of inner representations) still remain largely unknown. It is, for example, particularly difficult to reconcile the well known fact that ANNs achieve remarkable levels of generalization also in con- ditions of severe over-parametrization. In our work, we explore a recent network compression technique, called Iterative Magnitude Pruning (IMP), and apply it to convolutional neural networks (CNN). The pruned and unpruned models are compared layer-wise with Canonical Correlation Analysis (CCA). Our results show a high similarity between layers of pruned and unpruned CNNs in the first convolutional layers and in the fully-connected layer, while for the intermediate convolutional layers the similarity is significantly lower. This suggests that, although in intermediate layers representation in pruned and unpruned networks is markedly different, in the last part the fully-connected layers act as pivots, producing not only similar performances but also similar representations of the data, despite the large difference in the number of parameters involved.
- Toward the Application of Reinforcement Learning to the Intensity Control of a Seeded Free-Electron LaserNiky Bruchon, Gianfranco Fenu, Giulio Gaio, Marco Lonza, Felice Andrea Pellegrino, and Erica SalvatoIn 2019 23rd International Conference on Mechatronics Technology (ICMT), Oct 2019
The optimization of particle accelerators is a challenging task, and many different approaches have been proposed in years, to obtain an optimal tuning of the plant and to keep it optimally tuned despite drifts or disturbances. Indeed, the classical model-free approaches (such as Gradient Ascent or Extremum Seeking algorithms) have intrinsic limitations. To overcome those limitations, Machine Learning techniques, in particular, the Reinforcement Learning, are attracting more and more attention in the particle accelerator community. The purpose of this paper is to apply a Reinforcement Learning model-free approach to the alignment of a seed laser, based on a rather general target function depending on the laser trajectory. The study focuses on the alignment of the lasers at FERMI, the free-electron laser facility at Elettra Sincrotrone Trieste. In particular, we employ Q-learning with linear function approximation and report experimental results obtained in two setups, which are the actual setups where the final application has to be deployed. Despite the simplicity of the approach, we report satisfactory preliminary results, that represent the first step toward a fully automatic procedure for seed laser to the electron beam. Such a superimposition is, at present, performed manually.
- A Combined Support Vector Machine and Support Vector Representation Machine Method for Production ControlAntonio Acernese, Carmen Del Vecchio, Gianfranco Fenu, Luigi Glielmo, and Felice Andrea PellegrinoIn 2019 18th European Control Conference (ECC), Jun 2019
Machine learning techniques have been widely applied to production processes with the aim of improving product quality, supporting decision-making, or implementing process diagnostics. These techniques proved particularly useful in manufacturing industry where huge variety of heterogeneous data, related to different production processes, can be gathered and recorded but where traditional models fail due to the complexity of the production process. In this study, we describe a novel Machine Learning methodology to associate some product attributes (either defects or desirable qualities) to process parameters. Namely we combined Support Vector Ma- chine (SVM) and the Support Vector Representation Machine (SVRM) for instance ranking. The combination of SVM and SVRM guarantees a high flexibility in modeling the decision surfaces (thanks to the kernels) while containing overfitting (thanks to the principle of margin maximization). Thus this method is well suited for modeling unknown, possibly complex relationships, that may not be captured by simple handcrafted models. We applied our method to production data of an investment casting industry placed in South Italy. We obtain an instance ranking that may be used to infer proper values of process parameter set-points.
- Clinical Decision Support Using Colored Petri Nets: a Case Study on Cancer Infusion TherapyFrancesca Cairoli, Gianfranco Fenu, and Felice Andrea PellegrinoIn 2019 6th International Conference on Control, Decision and Information Technologies (CoDIT), Apr 2019
We consider a drug infusion scenario in which a drug is delivered through an infusion pump to a patient, whose vital parameters are monitored via a bedside monitor. Drug infusion therapies are based on clinical protocols that are drug-specific and very diversified. The burden of their proper application on several patients lies, most of the times, on nursing staff alone. With the aim of making the choices safe and prompt and limiting human errors, we build a system that suggests the proper action based on the protocol and the status of the patient. Given the high variability of protocols, it is important to choose a flexible structure. We choose Hierarchical Colored Petri Nets (HCPN), a mathematical formalism for describing discrete event dynamic systems, which is, in fact, modular, expressive and admits a graphic representation. Cancer infusion therapy is the case study considered, as that clinical scenario is likely to become critical from a staff/patient ratio point of view, since the number of patients is continuously growing.
- Model Predictive Control of glucose concentration based on Signal Temporal Logic specificationsFrancesca Cairoli, Gianfranco Fenu, Felice Andrea Pellegrino, and Erica SalvatoIn 2019 6th International Conference on Control, Decision and Information Technologies (CoDIT), Apr 2019
Insulin is a peptide hormone produced by the pancreas to regulate the cells intake of glucose in the blood. Type 1 diabetes compromises this particular capacity of the pancreas. Patients with this disease inject insulin to regulate the level of glucose in the blood, thus reducing the risk of longterm complications. Artificial Pancreas (AP) is a wearable device developed to provide automatic delivery of insuline, allowing a potentially significant improvement in the quality of life of patients. In this paper we apply to the AP a Model Predictive Controller able to generate state trajectories that meet constraints expressed through Signal Temporal Logic (STL). Such a form of constraints is indeed appropriate for the AP, in which some requirements result in hard constraints (absolutely avoid hypoglycaemia) and some other in soft constraints (avoid a prolonged hyperglycaemia). We rely on the BluSTL toolbox, which allows to automatically generate controllers using STL specifications. We perform simulations on two different scenarios: an MPC controller that uses the same constraints as [1] and an MPC-STL controller in both deterministic and adversarial environment (robust control). We show that the soft constraints permitted by STL avoid unnecessary restriction, providing safe trajectories in correspondence of higher disturbance.
- Model-free tuning of plants with parasitic dynamicsFranco Blanchini, Gianfranco Fenu, Giulia Giordano, and Felice Andrea PellegrinoIn 2017 56th IEEE Conference on Decision and Control (CDC), 2017
We have recently considered the problem of tun- ing a static plant described by a differentiable input-output function, which is completely unknown, but whose Jacobian takes values in a known polytope of matrices: to drive the output to a given desired value, we have suggested an integral feedback scheme, whose convergence is ensured if the polytope of matrices is robustly full row rank. The suggested tuning scheme may fail in the presence of parasitic dynamics, which may destabilize the loop if the tuning action is too aggressive. Here we show that such tuning action can be applied to dynamic plants as well if it is sufficiently “slow”, a property that we can ensure by limiting the integral action. We provide robust bounds based on the exclusive knowledge of the largest time constant and of the matrix polytope to which the system Jacobian is known to belong. We also provide similar bounds in the presence of parasitic dynamics affecting the actuators.
- Discrete-Time Trials for Tuning without a ModelFranco Blanchini, Gianfranco Fenu, Giulia Giordano, and Felice Andrea PellegrinoIn Proceedings of the 20th IFAC World Congress, 2017
Given a static plant described by a differentiable input-output function, which is completely unknown, but whose Jacobian takes values in a known polytope in the matrix space, we consider the problem of tuning the output (i.e., driving the output to a desired value), by suitably choosing the input. To this aim, we assume to have at our disposal a discrete sequence of trials only, as it happens, for instance, when we iteratively run a software, with new input data at each iteration, in order to achieve the desired output value. In this paper we prove that, if the polytope is robustly non-singular (or has full row rank, in the general non-square case), then a suitable discrete-time tuning law drives the output to the desired point. The computation of the tuning law is based on a convex-optimisation problem to be solved on-line. An application example is proposed to show the effectiveness of the approach.
- Segmentation of Mosaic Images Based on Deformable Models Using Genetic AlgorithmsAlberto Bartoli, Gianfranco Fenu, Eric Medvet, Felice Andrea Pellegrino, and Nicola TimeusIn Smart Objects and Technologies for Social Good: Second International Conference, GOODTECHS 2016 Proceedings, 2017
Preservation and restoration of ancient mosaics is a crucial activity for the perpetuation of cultural heritage of many countries. Such an activity is usually based on manual procedures which are typically lengthy and costly. Digital imaging technologies have a great potential in this important application domain, from a number of points of view including smaller costs and much broader functionalities. In this work, we propose a mosaic-oriented image segmentation algorithm aimed at identifying automatically the tiles composing a mosaic based solely on an image of the mosaic itself. Our proposal consists of a Genetic Algorithm, in which we represent each candidate segmentation with a set of quadrangles whose shapes and positions are modified during an evolutionary search based on multi-objective optimization. We evaluate our proposal in detail on a set of real mosaics which differ in age and style. The results are highly promising and in line with the current state-of-the-art.
- Computer Vision for the Blind: A Comparison of Face Detectors in a Relevant ScenarioMarco De Marco, Gianfranco Fenu, Eric Medvet, and Felice Andrea PellegrinoIn Smart Objects and Technologies for Social Good: Second International Conference, GOODTECHS 2016, Venice, Italy, November 30 – December 1, 2016, Proceedings, 2017
Motivated by the aim of developing a vision-based system to assist the social interaction of blind persons, the performance of some face detectors are evaluated. The detectors are applied to manually annotated video sequences acquired by blind persons with a glass-mounted camera and a necklace-mounted one. The sequences are relevant to the specific application and demonstrate to be challenging for all the considered detectors. A further analysis is performed to reveal how the performance is affected by some features such as occlusion, rotations, size and position of the face within the frame.
- Computer Vision for the blind: a dataset for experiments on face detection and recognitionSergio Carrato, Stefano Marsi, Eric Medvet, Felice Andrea Pellegrino, Giovanni Ramponi, and Michele VittoriIn Proceedings of the 39th International Convention on Information and Communication Technology, Electronics and Microelectronics, 2016
We present a video dataset created for the needs of a research project that aims at developing vision-based techniques that assist the social interaction of a blind person. Two totally blind users have acquired the sequences, using at the same time a glasses-mounted camera and a necklace-mounted one. The video sequences were acquired in different environments, selecting conditions in which a user could be interested in detecting the presence of some of his/her acquaintances, to approach them in a most natural way. The sequences have been temporally cropped to extract video shots that, by inspection, were deemed to contain events valuable for the goals of the project. The shots are presently being annotated, and some preliminary experiments on face detection have been performed on the annotated data. We also present some observations about the specific application that is being considered.
- Plant tuning: A robust Lyapunov approachFranco Blanchini, Gianfranco Fenu, Giulia Giordano, and Felice Andrea PellegrinoIn Decision and Control (CDC), 2015 IEEE 54th Annual Conference on, 2015
We consider the problem of tuning the output of a static plant whose model is unknown, under the only information that the input–output function is monotonic in each component or, more in general, that its Jacobian belongs to a known polytope of matrices. As a main result, we show that, if the polytope is robustly non–singular (or has full rank, in the non–square case), then a suitable tuning scheme drives the output to a desired point. The proof is based on the application of a well known theorem concerning the existence of a saddle point for a min–max zero–sum game. Some application examples are suggested.
- On the Assessment of Segmentation Methods for Images of MosaicsGianfranco Fenu, Nikita Jain, Eric Medvet, Felice Andrea Pellegrino, and Myriam Pilutti NamerIn Proceedings of 10th International Conference on Computer Vision Theory and Applications VISAPP 2015, 2015
The present paper deals with automatic segmentation of mosaics, whose aim is obtaining a digital representation of the mosaic where the shape of each tile is recovered. This is an important step, for instance, for preserving ancient mosaics. By using a ground-truth consisting of a set of manually annotated mosaics, we objectively compare the performance of some existing recent segmentation methods, based on a simple error metric taking into account precision, recall and the error on the number of tiles. Moreover, we introduce some mosaic-specific hardness estimators (namely some indexes of how difficult is the task of segmenting a particular mosaic image). The results show that the only segmentation algorithm specifically designed for mosaics performs better than the general purpose algorithms. However, the problem of segmentation of mosaics appears still partially unresolved and further work is needed for exploiting the specificity of mosaics in designing new segmentation algorithms.
- Image Processing Issues in a Social Assistive System for the BlindMargherita Bonetto, Sergio Carrato, Gianfranco Fenu, Eric Medvet, Enzo Mumolo, Felice Andrea Pellegrino, and Giovanni RamponiIn Image and Signal Processing and Analysis (ISPA), 2015 9th International Symposium on, 2015
We systematically analyse the design of the low- level vision components of a real-time system able to help a blind person in his/her social interactions. We focus on the acquisition and processing of the video sequences that are acquired by a wearable sensor (a smartphone camera or a Webcam) for the detection of faces in the scene. We review some classical and some very recent techniques that seem appropriate to the requirements of our goal.
- Inverse kinematics by means of convex programming: some developmentsFranco Blanchini, Gianfranco Fenu, Giulia Giordano, and Felice Andrea PellegrinoIn Proceedings of the 11th IEEE International Conference on Automation Science and Engineering (CASE 2015), 2015
A novel approach to the problem of inverse kinematics for redundant manipulators has been recently introduced: by considering the joints as point masses in a fictitious gravity field, and by adding proper constraints to take into account the length of the links, the kinematic inversion may be cast as a convex programming problem. Such a problem can be solved in an efficient way and may be easily modified to include constraints due to obstacles. Here we present further developments of the idea. In particular, for the case of planar robots, we show (i) how to impose hard constraints on the joint angles while preserving the convexity of the problem, and (ii) how to add constraints due to objects (for instance, a load carried by the robot) that are rigidly attached to some part of the robot.
- Evaluation of features for automatic detection of cell nuclei in fluorescence microscopy imagesPaolo Fabris, Walter Vanzella, and Felice Andrea PellegrinoIn Image and Signal Processing and Analysis (ISPA), 2013 8th International Symposium on, 2013
The problem of detecting cell nuclei in images may be faced by means of a segmentation the neighbourhood of candidate nuclei, followed by a binary classification step. Important for the latter step is the choice of the descriptors (features) to be extracted from the neighbourhood and used by the classifier. In the present paper, based on a large set of manually labelled samples, we evaluate several of such descriptors combined with some common type of support vector machines. We show that equipping the detection algorithm with the best combination of features/classifier leads to a performance comparable to human labelling by experts.
- Experimental setup and methodology for automatic quality assessment of intraocular lensesAndrea Cigaina, Gianfranco Fenu, Felice Andrea Pellegrino, Paolo Sirotti, Silvia Rinaldi, and Daniele TognettoIn Image and Signal Processing and Analysis (ISPA), 2013 8th International Symposium on, 2013
Intraocular lenses (IOLs) are widely used in cataract surgery. There is a variety of IOLs depending on lens material, optical and mechanical design. Evaluation of the visual performance obtainable with intraocular lenses is important for objective comparison between IOL models. Indeed, the visual performance of implanted lenses has a deep impact on the life of the subjects, having medical, social and economical consequences. The purpose of the present paper is twofold: first, we propose an experimental setup, consisting of an actuated opto–electro– mechanical eye capable of simulating different refractive and optical conditions. Second, we suggest to employ some well– known image quality assessment tools for evaluating the visual performance of different IOLs (or of the same IOL in different conditions) based on the comparison between the original image and that produced by the artificial eye equipped with the IOL. The objective and repeatable results obtained, are in accor- dance with subjective evaluation (a time–consuming procedure employed in previous works).
- Application of Hamiltonian dynamics to manipulator control in constrained workspaceDaniele Casagrande, Gianfranco Fenu, Felice Andrea Pellegrino, and Alessandro AstolfiIn Proceedings of the 52nd IEEE Conference on Decision and Control, 2013
We describe a methodology to plan the trajectory of a robotic manipulator moving in an n-dimensional Euclidean space. The robot has to reach a target avoiding obstacles and is allowed to move in a constrained space. We show that if the position of the target and of the obstacles in known a priori, then a Hamiltonian function can be constructed and used to define the trajectory. A simulation concerning a real three-dimensional manipulator is finally reported.
- Quality of Images with Multifocal IOLsDaniele Tognetto, Silvia Rinaldi, Claudia Papagno, Gianfranco Fenu, Felice Andrea Pellegrino, and Paolo SirottiIn ASCRS-ASOA Symposium & Congress, 2013
- Quality of Images With Premium IOLsDaniele Tognetto, Silvia Rinaldi, Claudia Papagno, Gianfranco Fenu, Felice Andrea Pellegrino, and Paolo SirottiIn ASCRS Symposium on Cataract, IOL and Refractive Surgery, 2013
To evaluate the quality of images obtained with different type of premium intraocular lenses (IOLs) by developing an optoelectronic test bench including an artificial eye. An experimental opto-electronic test bench has been realized following the Gullstrand eye model and ISO 11979-2 norm. The system includes an electro-opto-mechanical device that simulates the optical and geometric characteristics of the human eye. The PMMA artificial cornea has been realized with optical and geometric properties similar to those of a real cornea. A cylindrical lens is placed in front of the artificial cornea to simulate corneal astigmatism. Toric IOL were analized evaluating the Modulation Transfer Function (MTF) and the quality of different images. A relevant software was realized to capture, store images and evaluate their quality in accordance with established criteria. Images transmitted through an astigmatic cornea are corrected by the toric IOL as much as the axis of the lens correspond to the axis of the astigmatism thus improving the transmission of different spatial frequencies and the quality of test images. Pupil diameter affects the quality of images especially when a residual refractive error is induced. This opto-electronic test bench allows to determine the quality of images transmitted by premium intraocular lenses in different refractive and optical conditions. Toric IOL have been studied demonstrating their ability to correct the corneal astigmatism. This system allows to evaluate different criteria besides the MTF, such as the evaluation of sharpness or the distortion of local image patterns or even to formulate new objective image evaluating criteria.
- Hamiltonian dynamics for robot navigationDaniele Casagrande, Felice Andrea Pellegrino, and Alessandro AstolfiIn The 9th IEEE International Conference on Control and Automation (ICCA), 2011
We describe a methodology to plan the trajectory of a robot moving in a two-dimensional space. The robot has to perform the task of reaching a target avoiding obstacles. We show that if the position of the target and of the obstacles in known a priori, then a suitable Hamiltonian function can be constructed and used to define the trajectory. We consider both the static case, namely the case in which both the target and the obstacles are fixed, and the dynamic case, namely the case in which the target or the obstacles move. We prove that in both cases the target can be reached in finite time. The paper is enriched by several examples that accompany the discussion.
- Trajectory clustering by means of Earth Mover’s DistanceFrancesca Boem, Felice Andrea Pellegrino, Gianfranco Fenu, and Thomas ParisiniIn Proceedings of IFAC World Congress, Milano, Italy, 2011
We propose a method for trajectory classification based on a general cluster-based methodology, that can be used both off-line in an unsupervised fashion, both on-line, classifying new trajectories or part of them. We use the Earth Mover’s Distance (EMD) and we adapt it in order to employ it as a tool for trajectory clustering. We propose a novel effective method to identify the clusters’ representatives by means of the p−median location problem. This methodology is able to manage different length and noisy trajectories and takes velocity profiles and stops into account. We discuss the experimental results and we compare our approach with other trajectory clustering methods.
- Multi-feature trajectory clustering using Earth Mover’s DistanceFrancesca Boem, Felice Andrea Pellegrino, Gianfranco Fenu, and Thomas ParisiniIn Proceedings of the IEEE Conference on Automation Science and Engineering, Trieste, Italy, 2011
We present new results in trajectory clustering, obtained by extending a recent methodology based on Earth Mover’s Distance (EMD). The EMD can be adapted as a tool for trajectory clustering, taking advantage of an effective method for identifying the clusters’ representatives by means of the p−median location problem. This methodology can be used either in an unsupervised fashion, or on-line, classifying new trajectories or part of them; it is able to manage different length and noisy trajectories, occlusions and takes velocity profiles and stops into account. We extend our previous work by taking into account other features besides the spatial locations, in particular the direction of movement in correspondence of each trajectory point. We discuss the simulation results and we compare our approach with another trajectory clustering method.
- Fast and Accurate Object Detection by Means of Recursive Monomial Feature Elimination and Cascade of SVMLorenzo Dal Col, and Felice Andrea PellegrinoIn Proceedings of the IEEE Conference on Automation Science and Engineering, Trieste, Italy, 2011
Support Vector Machines (SVMs) are an established tool for pattern recognition. However, their application to real–time object detection (such as detection of objects in each frame of a video stream) is limited due to the relatively high computational cost. Speed is indeed crucial in such applications. Motivated by a practical problem (hand detection), we show how second–degree polynomial SVMs in their primal formulation, along with a recursive elimination of monomial features and a cascade architecture can lead to a fast and accurate classifier. For the considered hand detection problem we obtain a speed–up factor of 1600 with comparable classification performance with respect to a single, unreduced SVM.
- Approximate off-line receding horizon control of constrained nonlinear discrete-time systems: Smooth approximation of the control lawGilberto Pin, Marco Filippo, Felice Andrea Pellegrino, Gianfranco Fenu, and Thomas ParisiniIn American Control Conference (ACC), 2010, 2010
In this work, the off-line approximation of state-feedback nonlinear model predictive control laws by means of smooth functions of the state is addressed. The idea is to investigate how the approximation errors affect the stability of the closed-loop system, in order to derive suitable bounds which have to be fulfilled by the approximating function. This analysis allows to conveniently set up the characteristic parameters of some techniques such as Neural Networks which can be used to implement the control law, in order to render the system Input-to-State Practically Stable while satisfying, in addition, hard constraints on the trajectories; both the amount of data storage and the computational time result strongly reduced with respect to Nearest Neighbor or Set Membership approaches, which have been recently proposed to obtain effective off-line approximation of nonlinear MPC. The provided simulations confirm the validity of the method.
- Approximate off-line receding horizon control of constrained nonlinear discrete-time systemsGilberto Pin, Marco Filippo, Felice Andrea Pellegrino, and Thomas ParisiniIn Proc. of the European Control Conference, 2009
The present paper concerns the design of approximate off-line model predictive control laws for nonlinear discrete-time systems subject to hard constraints on state and input variables. The possibility to obtain an approximate receding horizon control law by performing off-line optimization, leads to a dramatic reduction of the real-time computational complexity with respect to on-line algorithms, and allows the application of the developed control technique to plants with fast dynamics, that require small sampling periods. The main feature of the proposed approximation scheme consists in the possibility to cope with possibly discontinuous state-feedback control laws, while guaranteeing the fulfillment of hard constraints on state and input variables despite the perturbations due to the use of an approximate controller. Finally, the resulting closed-loop system is shown to be input-to-state-stable with respect to the approximation-induced perturbations.
- High-gain adaptive control: A derivative-based approachFranco Blanchini, Thomas Parisini, Felice Andrea Pellegrino, and Gilberto PinIn Proceedings of the 47th IEEE Conference on Decision and Control, 2008
We propose an adaptive scheme which is a coun- terpart of existing high gain control techniques based on control Lyapunov functions. Given a control Lyapunov function, the main idea is that of tuning the feedback gain according to a suitably–chosen Lyapunov time–derivative. The control gain is not monotonically non–decreasing as in existing techniques, but it is increased or decreased depending on the imposed derivative, thus avoiding the well-known issue of actuator over– exploitation. We are able to show robust convergence of the proposed adaptive control scheme as well as other interesting properties. For instance, it is possible to guarantee an a–priori given upper bound for the transient mode of behavior during adaptation. Furthermore, if the control Lyapunov function is designed based on an optimal control problem, then the control action is nominally optimal, precisely it yields the optimal trajectory for any initial condition, if the actual plant matches the nominal system.
- Relatively optimal control: a static piecewise-affine solutionFranco Blanchini, and Felice Andrea PellegrinoIn Proceedings of the 46th IEEE Conference on Decision and Control, 2007
A relatively optimal control is a stabilizing controller that, without initialization nor feedforwarding and tracking the optimal trajectory, produces the optimal (constrained) behavior for the nominal initial condition of the plant. In a previous work, for discrete–time linear systems, we presented a linear dynamic relatively optimal control. Here we provide a static solution, namely a dead–beat piecewise affine state–feedback controller based on a suitable partition of the state space into polyhedral sets. The vertices of the polyhedra are the states of the optimal trajectory, hence a bound for the complexity of the controller is known in advance. We also show how to obtain a controller that is not dead–beat by removing the zero terminal constraint while guaranteeing stability.
- Compensator blending: a new tool for multiobjective designFranco Blanchini, Patrizio Colaneri, and Felice Andrea PellegrinoIn Proceedings of the 5th IFAC Symposium on Robust Control Design, 2006
In this paper we show that (under some input matrix rank conditions) there exists a single compensator which achieves simultaneously the performances of r ≤ n (the system order) given static state feedback (local) compensators. The compensator, whose order is r(n − 1), is then capable of matching the r (possibly different) optimality criteria defined for each input-output pair. An explicit and easy construction procedure (we refer to such procedure as “compensator blending”) is provided. We also consider the dual version of the problem, precisely, we show how to achieve simultaneous optimality by blending a set of given filters.
- Relatively Optimal Control: the Static SolutionFranco Blanchini, and Felice Andrea PellegrinoIn Proceedings of the 16th IFAC World Congress, 2005
A relatively optimal control is a stabilizing controller that, without initialization nor feedforwarding and tracking the optimal trajectory, produces the optimal (constrained) behavior for the nominal initial condition of the plant. In a previous work, a linear dynamic relatively optimal control, for discrete–time linear systems, was presented. Here a static solution is shown, namely a dead–beat piecewise affine state–feedback controller based on a suitable partition of the state space into polyhedral sets. The vertices of the polyhedrons are the states of the optimal trajectory, hence a bound for the complexity of the controller is known in advance. It is also shown how to obtain a controller that is not dead–beat by removing the zero terminal constraint while guaranteeing stability. Finally, the proposed static compensator is compared with the dynamic one.
- Enhancing controller performance for robot positioning in a constrained environmentFranco Blanchini, Felice Andrea Pellegrino, Stefano Miani, and Bart ArkelIn Proceedings of the 43rd IEEE Conference on Decision and Control, 2004
The paper considers a novel technique for manip- ulator motion in constrained environment due to the presence of obstacles. The basic problem is that of avoiding collisions of the manipulator with the obstacles. The main idea is to cover the free space (i.e. the points of the configurations space in which no collisions are possible) by a connected family of polyhedral sets which are controlled–invariant. Each of these polyhedral regions includes some crossing points to the confining regions. The tracking control is hierarchically struc- tured. A high–level controller establishes a connected chain of regions to be crossed to reach the one in which the reference is included. A low–level control solves the problem of tracking, within a region, the crossing point to the next confining region and, eventually, tracking the reference whenever it is included in the current one. The scheme assures that the reference is asymptotically tracked and that the transient trajectory is completely included in the admissible configuration space. A connection graph associated with the cluster of regions, and the high–level control is achieved by solving a minimum–path problem. As far as the low–level control is concerned, we consider both speed–control and torque–control. We propose two types of controllers. The first type is based on a linear stabilizing feedback which is suitably adapted to achieve a local tracking controller. Such a controller is computed by the plane representation of the sets which is more natural and useful then the vertex representation considered in previous work. The second is a speed–saturated type of controller which considerably improves the performance of linear–based control laws. Both these controllers have a speed–control and torque–control version. Experimental results on a laboratory Cartesian robot are provided.
- Relatively optimal control with characteristic polynomial assignmentFranco Blanchini, and Felice Andrea PellegrinoIn Proceedings of the 43rd IEEE Conference on Decision and Control, 2004
A relatively optimal control is a stabilizing controller such that, if initialized at its zero state, produces the optimal (constrained) behavior for the nominal initial condition of the plant (without feedforwarding and tracking the optimal trajectory). In a previous work we have shown that one of such controllers is linear, dead-beat, and its order is equal to the length of the horizon minus the plant order, thus of complexity which is known a-priori. In this work we remove the assumption of zero terminal state and we show how to assign an arbitrary closed-loop characteristic stable polynomial to the plant (an explicit solution is provided.) We also show how to choose the characteristic polynomial in such a way that the constraints (which are enforced on a finite horizon) can be globally or ultimately satisfied.
- Relatively optimal control and its linear implementationFranco Blanchini, and Felice Andrea PellegrinoIn Proceedings of the 42nd IEEE Conference on Decision and Control, 2003
Motivated by the fact that determining a feedback solution for the optimal control problem under constraints is a hard task we introduce the concept of relative optimality, roughly optimality for a specific (nominal) plant initial condition. We consider a generic discrete-time finite-horizon constrained optimal control problem for linear systems, and we seek for a state feedback (possibly dynamic) controller. As a fundamental requirement, we do not admit preactions or controller-state initialization based on the plant initial state and we assume our controller to be time-invariant. In particular, we do not consider controllers simply achieved by the feedforward and tracking of the optimal trajectory. A relatively optimal control is a stabilizing controller such that, if initialized at its zero state, produces the optimal (constrained) trajectory for the nominal initial condition of the plant. We show that one of such controllers is linear, dead- beat, and its order is equal to the length of the horizon minus the plant order, thus of complexity which is known a-priori.
- Suboptimal receding horizon control for continuous-time systemsFranco Blanchini, Stefano Miani, and Felice Andrea PellegrinoIn Proceedings of the 41st IEEE Conference on Decision and Control, 2002
In this paper, a continuous–time optimal control problem is approached in a sub–optimal way by introducing the concept of suboptimal value function, which is any function satisfying the Hamilton–Jacobi–Bellman inequality. It is shown that as long as the Euler Approximating System (EAS) of a given continuous–time plant admits a positive definite convex suboptimal value function, it is possibile to determine a stabilizing control for the continuous–time system whose cost not only converges to the optimal, but it is also upper bounded by the discrete–time cost no matter how the “discretization time parameter” is chosen.
- How the spatial filters of area V1 can be used for a nearly ideal edge detectionFelice Andrea Pellegrino, Walter Vanzella, and Vincent TorreIn Proceedings of the 2nd International Workshop on Biological Motivated Computer Vision, 2002
The present manuscript aims to address and possibly solve three classical problems of edge detection: i – the detection of all step edges from a fine to a coarse scale; ii – the detection of thin bars, i.e. of roof edges; iii – the detection of corners and trihedral junctions. The proposed solution of these problems combines an extensive spatial filtering, inspired by the receptive field properties of neurons in the visual area V1, with classical methods of Computer Vision (Morrone & Burr 1988; Lindeberg 1998; Kovesi 1999) and newly developed algorithms. Step edges are computed by extracting local maxima from the energy summed over a large bank of odd filters of different scale and direction. Thin roof edges are computed by considering maxima of the energy summed over narrow odd and even filters along the direction of maximal response. Junctions are precisely detected by an appropriate combination of the output of directional filters. Detected roof edges are cleaned by using a regularization procedure and are combined with step edges and junctions in a Mumford-Shah type functional with self adaptive parameters, providing a nearly ideal edge detection and segmentation.
Book Chapters
- Investigating Similarity Metrics for Convolutional Neural Networks in the Case of Unstructured PruningAlessio Ansuini, Eric Medvet, Felice Andrea Pellegrino, and Marco ZullichIn Pattern Recognition Applications and Methods, 2020
Deep Neural Networks (DNNs) are essential tools of modern science and technology. The current lack of explainability of their inner workings and of principled ways to tame their architectural complexity triggered a lot of research in recent years. There is hope that, by making sense of representations in their hidden layers, we could collect insights on how to reduce model complexity—without performance degradation—by pruning useless connections. It is natural then to ask the following question: how similar are representations in pruned and unpruned models? Even small insights could help in finding principled ways to design good lightweight models, enabling significant savings of computation, memory, time and energy. In this work, we investigate empirically this problem on a wide spectrum of similarity measures, network architectures and datasets. We find that the results depend critically on the similarity measure used and we discuss briefly the origin of these differences, concluding that further investigations are required in order to make substantial advances.
- Towards More Natural Social Interactions of Visually Impaired PersonsSergio Carrato, Gianfranco Fenu, Eric Medvet, Enzo Mumolo, Felice Andrea Pellegrino, and Giovanni RamponiIn Advanced Concepts for Intelligent Vision Systems SE - 63, 2015
We review recent computer vision techniques with reference to the specific goal of assisting the social interactions of a person affected by very severe visual impairment or by total blindness. We consider a scenario in which a sequence of images is acquired and processed by a wearable device, and we focus on the basic tasks of detecting and recognizing people and their facial expression. We review some methodologies of Visual Domain Adaptation that could be employed to adapt existing classification strategies to the specific scenario. We also consider other sources of information that could be exploited to improve the performance of the system.
Technical Reports
- Some recent results in Trajectory ClusteringFrancesca Boem, Felice Andrea Pellegrino, Gianfranco Fenu, and Thomas Parisini2011
We illustrate recent and ongoing research on trajectory clustering. Most of the results have been already published, but some new achievements are present. The main idea is that of using an adapted formulation of the Earth Mover’s Distance (EMD) as a tool for trajectory clustering. We proposed a novel effective method to identify the clusters’ representatives by means of the p−median location problem. We extended our first work by taking into account other features besides the spatial locations.
Miscellaneous
- Metodo per il rilevamento del traffico pedonale in uno spazioGiovanni Longo, Felice Andrea Pellegrino, Cristian Giacomini, Gianfranco Fenu, and Andrea Assalone2018
-
- Procedimento per il riconoscimento ed il conteggio di particelle e relativa apparecchiatura, 0001390204Felice Andrea Pellegrino, and Walter Vanzella2008
- Proprietà strutturali dei sistemi lineari invariantiFranco Blanchini, and Felice Andrea Pellegrino2007
- Modulo di Controlli Automatici I, EserciziarioFelice Andrea Pellegrino2004
Il presente Eserciziario è stato realizzato come supporto al corso di Controllo Automatici I tenuto dal prof. Stefano Miani nell’ambito del Corso di Laurea in Ingegneria Meccanica, Università degli Studi di Udine, sede di Pordenone. Gli argomenti trattati seguono fedelmente il programma di tale insegnamento. Ringrazio di cuore chi vorrà segnalarmi eventuali errori e/o imprecisioni.
- Analisi dei sistemi lineariFranco Blanchini, and Felice Andrea Pellegrino2004
Questo testo ‘e stato pensato e scritto per studenti del corso di Controlli Automatici I della Facolta‘ di Ingegneria dell’Universita‘ di Udine, Corso di Laurea in Ingegneria Elettronica e Ingegneria Gestionale dell’Informazione e come testo riassuntivo dei concetti basilari per i corsi di Controlli Automatici II e il corso integrato di Teoria dei Sistemi I e II dei medesimi corsi di Laurea. Lo spirito della stesura non ‘e stato quello di creare un nuovo libro ma molto piu‘ semplicemente degli appunti che riassumano in modo breve e schematico i concetti presentati nel corso. Questa dispensa (anche se ‘e un’esagerazione chiamiamola cos‘ı) non ha alcuna pretesa di completezza e non deve in nessun modo essere pensata come sostitutiva di un buon libro di testo dal quale non si puo‘ prescindere. Sono disponibili molti libri di testo di Controlli Automatici di alta qualita‘. Apprendere tramite libri ad ampia diffusione ‘e importante perch ́e permette a studenti di diversa provenienza di acquisire un linguaggio comune. Quindi i libri consigliati sono da ritenersi fondamentali per lo studio dei corsi di Controlli Automatici. Il presente testo deve essere dunque considerato come un riassunto schematico dei concetti presentati nel corso ed una integrazione del libro di testo.
Theses
- Constrained and Optimal ControlFelice Andrea PellegrinoUniversity of Udine, Udine, Italy, 2005
Constraint fulfillment and optimality are two important aspects of most real–world con- trol problems. Constraints typically arise from safety, physical or performance limita- tions and must be taken into account in the design stage, especially in those cases when constraint violation may have serious consequences. As a matter of fact, the control of systems subject to input or state constraints is a difficult task, being the problem in- herently nonlinear. The problem becomes still harder if optimality is also pursued. The most popular approaches to optimality for linear systems (namely LQR, LQG, H∞, L1) do not take into account constraints. The more general tools of dynamic programming and Pontryagin minimum principle are not useful in most cases, the former involving an- alytical and numerical difficulties, the latter being suitable for open–loop solutions only. The receding horizon approach allows for dealing with constraints in a suboptimal (and sometimes optimal) way but it requires solving on–line an optimization problem at each time step. In this thesis we review some of the basic techniques for constrained optimal control focusing our attention to those that are the most effective from a computational point of view. We also present an original contribution, namely the relatively optimal control (ROC) that is a stabilizing feedback control, synthesized starting from a given (or a set of given) open–loop optimal trajectories, that guarantees optimality for a given (or a set of given) initial conditions. The ROC is based on relaxing the requirement of obtaining a controller that is optimal for all possible initial conditions; on the contrary, we require optimality only from some initial conditions that are of interest for the con- sidered system. For linear systems, we provide a linear dynamic as well as a piecewise affine static solution and we develop various extensions of the basic idea.
-