ABSTRACT. Symbolic Regression (SR) is a machine learning approach developed for the automated identification of mathematical equations that accurately capture the relationships between input and output features within the experimental dataset. This method is capable of creating interpretable models while incorporating existing knowledge into the system. This paper addresses a problem in the development of interpretable Soft Sensors (SS) for industrial applications using SR. The challenge arises from the need to increase the dimensionality of the problem to capture the system dynamics, which often leads to a significant degradation in SR performance. Existing literature has highlighted this problem and proposed some solutions, such as employing Recurrent Neural Networks (RNN) instead of Genetic Programming (GP) in the SR procedure or applying Deep Learning (DL) techniques to reduce the input space. In this paper, we present a novel approach to develop interpretable SSs for industrial processes that involve the use of DL to encode the system dynamics. This effectively reduces the input space and supports the SR process without compromising the interpretability of the final solution.

ABSTRACT. Traditionally based on convexity, multi-agent decision-making models can hardly handle scenarios where agents' cost functions defy this assumption, which is specifically required to ensure the existence of several equilibrium concepts. More recently, the advent of machine learning (ML), with its inherent non-convexity, has changed the conventional approach of pursuing convexity at all costs. This paper explores and integrates the robustness of game theoretic frameworks in managing conflicts among agents with the capacity of ML approaches, such as deep neural networks (DNNs), to capture complex agent behaviors. Specifically, we employ feed-forward DNNs to characterize agents' best response actions rather than modeling their goals with convex functions. We introduce a technical assumption on the weight of the DNN to establish the existence of Nash equilibria. Finally, we demonstrate the practical application of our framework to a noncooperative community of smart energy users under a dynamic time-of-use energy pricing scheme.

ABSTRACT. We present a sample-efficient approach to certify barrier functions of discrete-time dynamical systems. Our approach divides the 0-sublevel set of the candidate barrier function in multiple segments, and derives sampling resolution bounds in each segment to verify the barrier function with a user-specified decay rate.

ABSTRACT. This abstract presents a novel result for continuous-time smooth monotone systems: if the system is nonexpansive w.r.t. a diagonally weighted infinity norm and if there is at least one equilibrium point, then all trajectories converge to one of them. Differently from the current literature trend, the system is not required to be contractive but merely nonexpansive, thus allowing for multiple equilibrium points. Sufficient conditions on the vector field are provided to verify nonexpansiveness, which are then applied in the context of Recurrent Neural Networks (RNNs), thus identifying a class monotone RNNs whose trajectories are convergent even though their dynamics is not contractive.

ABSTRACT. This abstract summarises our research on Recurrent Neural Network (RNN) based Model Predictive Control (MPC). First the advantages of the use of RNN models in MPC are detailed. Then it is reported how the key issues of stability, constraint satisfaction and offset-free tracking have been solved in the MPC design, based on different classes of RNN.

ABSTRACT. We present a modeling and control design framework for modeling and influencing the drivers’ decisions in highway scenarios using one or more vehicles as actuators. Our approach relies on a driver’s decision-making model that is used to design a scenario-tree model predictive controller, which calculates acceleration and lane change commands for a set of controlled vehicles. We illustrate our modeling and control framework in a two-lane highway example, with two vehicles, one autonomous and one human-driven. Results from numerical simulations demonstrate how our approach can efficiently influence the lane changes of one vehicle using the other as a control actuator.

ABSTRACT. We present a novel approach to address the challenge of stop-and-go wave mitigation in congested vehicular traffic scenarios. The proposed solution involves equipping human-driven vehicles with a controller that can effectively assist the driver by merging human input with the underlying automation input through arbitration. Specifically, our approach integrates a convex combination of human and automation inputs within the controller through a continuous and derivable sharing function. This integration allows for the fusion of human decision-making capabilities with automation's perception of the environment. We provide extensive microscopic simulation results to demonstrate the effectiveness of the proposed approach. In addition, some theoretical guarantees are established for both the stability of individual vehicles and the string stability.

ABSTRACT. The behavior of cyclists when choosing the path to follow along a road network is not uniform. Some of them are mostly interested in minimizing the travelled distance, but some others may also take into account other features such as safety of the roads or pollution. Individuating the different groups of users, estimating the numerical consistency of each of these groups, and reporting the weights assigned by each group to different characteristics of the road network, is quite relevant. Indeed, when decision makers need to assign some budget for infrastructural interventions, they need to know the impact of their decisions, and this is strictly related to the way users perceive different features of the road network. In this paper, we propose an optimization approach to detect the weights assigned to different road features by various user groups, leveraging knowledge of the true paths followed by them, accessible, for example, through data collected by bike-sharing services

ABSTRACT. Urban autonomous driving has the potential to enhance both safety and efficiency of transportation in environments also in complex traffic conditions. However, new services and approaches are necessary to manage Autonomous Vehicles in the real traffic. This paper introduces a novel approach to optimize routing in the urban settings by Deep Reinforcement Learning (DRL) techniques. A modular DRL architecture is proposed to obtain a route able to minimize the length of the paths, minimize the number of turns during the travel and select the dedicated lanes.

ABSTRACT. In this abstract, we propose a route planning strategy for heterogeneous mobile robots in Precision Agriculture (PA) settings. Given a set of agricultural tasks to be performed at specific locations, we formulate a multi-Steiner Traveling Salesman Problem (TSP) to define the optimal assignment of these tasks to the robots as well as the respective optimal paths to be followed. Costs for travelling from one location to another, for maneuvering and for executing the task as well as limited energy capacity of the robots are considered. In addition, we propose a sub-optimal formulation to mitigate the computational complexity. The approach is validated in a simulated orchard with three heterogeneous aerial vehicles performing inspection tasks.

ABSTRACT. This abstract presents an heuristic online optimization algorithm, based on gossiping, to solve a workforce routing, task assignment, and scheduling problem with privacy by design, drawing inspiration from multi-vehicle routing problems. A real case study is considered, which involves a large number of technicians tasked with refurbishing and repairing a large number of photo booth machines spread across a wide geographic area, spanning a country. The heuristic can be used both offline and online to accommodate delays and unforeseen impediments encountered by technicians. The objective is to maximize enterprise profit by effectively managing the workforce. The proposed method inherently safeguards the privacy of real-time geolocation data for the entire workforce, ensuring it remains undisclosed and inaccessible to the company's ICT infrastructure. Numerical simulation - on real data supplied by DEDEM S.p.A. - demonstrate the performance of the proposed heuristic in terms of expected net profit for the company.

ABSTRACT. Autonomous vehicle racing is becoming increasingly popular in research as a platform to advance autonomous driving technologies. Recently, graph-based trajectory planners have been developed for autonomous racing with obstacle avoidance. These planners use a graph of motion primitives, which are trajectories connecting pairs of waypoints. The existing methods to compute motion primitives are optimization-based, geometric curve-based, and learning-based, each with its drawbacks: optimization methods are computationally heavy, geometric curves neglect the vehicle dynamics, and neural networks need large training sets to generalize well. This paper presents novel neural network-based motion primitives for autonomous racing, aimed at learning and generalizing the minimum-time vehicle trajectories.

ABSTRACT. The Serret-Frenet frame is widely used as an orthonormal basis of SO(3), but it cannot be applied to curves with zero-curvature arcs due to the frame indefiniteness. We discuss an alternative based on the Bishop frame and show the advantage both in simulation and with a robotic arm.

ABSTRACT. This contribution explores the application of Nonlinear Model Predictive Control (NMPC) techniques, based on the Pontryagin Minimum Principle, for a minimum-propellant autonomous rendezvous maneuver in a non-Keplerian Lunar orbits. The relative motion between the chaser and the target is described by the nonlinear dynamics of the circular restricted three body-problem, posing unique challenges due to the complex and unstable dynamics of near-rectilinear halo orbits. Key aspects of the proposed NMPC include trajectory optimization, maneuver planning, and real-time control, highlighting its ability to satisfy complex mission requirements, while ensuring safe and efficient spacecraft operations, in the presence of input and nonlinear/non-convex state constraints. The proposed formulation allows us to design minimum-propellant controller, whose optimal control signal results to be bang-bang in time. A case study based on the Artemis III mission - where the docking of the Orion spacecraft to the Gateway station is planned - is illustrated in order to demonstrate the efficiency of the proposed approach, showcasing its potential for enhancing target tracking accuracy, while reducing propellant consumption.

ABSTRACT. In mobile robotics and autonomous vehicles, path planning involves the computation of optimal trajectories while adhering to constraints like maximum curvature. The Three Point Dubins Problem (3PDP) is a special case where the path passes through three points with specified initial and final orientations. Solving this requires determining the angle at the middle point, which leads to a non-smooth and discontinuous solution. This work presents a numerical optimization approach that improves computational efficiency over brute force and other existing methods. Numerical tests demonstrate the algorithm's superior performance by comparing the number of iterations and computational time.

ABSTRACT. Clothoids are curves characterised by linear curvature. Their use is transversal in many applications ranging from motorway junctions to computer graphics passing through computer numerically controlled machines. Tracking these curves has an impact on efficiency in many fields. We present some preliminary results on a control system for the tracking of clothoids that relies on the so-called Internal Model Principle.

ABSTRACT. We consider the speed planning problem for a vehicle moving along an assigned trajectory, under maximum speed, tangential and lateral acceleration, and jerk constraints. The problem is a nonconvex one, where nonconvexity is due to jerk constraints. We propose a convex relaxation and show that the relaxation is exact under some assumptions. Also, we rewrite the relaxation as a Second Order Cone Programming (SOCP) problem. This has a relevant practical impact, since solvers for SOCP problems are quite efficient and allow solving large instances within tenths of a second. We conjecture that the convex relaxation is always exact, although we could not give a formal proof of this fact.

ABSTRACT. The development of algorithms for secure state estimation in vulnerable cyber-physical systems has been gaining attention in the last years. A consolidated assumption is that an adversary can tamper a relatively small number of sensors. In this paper, we propose a Lasso-based approach and we analyse its effectiveness. We theoretically derive conditions that guarantee successful attack/state recovery and we develop a sparse state observer. We compare the proposed methods to the state-of-the-art algorithm via numerical simulations.

ABSTRACT. This submission aims to present an interesting property of linear and time-invariant systems, namely negativizability: a pair (A,C) is negativizable if a gain matrix K exists such that A-KC is negative definite. Leveraging this property, we propose a novel fully distributed estimation and control scheme that allows each agent to locally perform the calculation of its estimation and control gain matrices. Notably, unlike other methodologies in the literature, our approach considers an interdependent setting that is natively integrable with the interconnected structure of cyber-physical systems. To conclude, we extended the proposed methodology to address the distributed estimation problem in the case of nonlinear systems.

ABSTRACT. For a reaction-diffusion equation with unknown right-hand side and non-local measurements subject to unknown constant measurement delay, we consider the nonlinear inverse problem of estimating the associated leading eigenvalues and measurement delay from a finite number of noisy measurements. We propose a reconstruction criterion and, for small enough noise intensity, prove existence and uniqueness of the desired approximation and derive closed-form expressions for the first-order condition numbers, as well as bounds for their asymptotic behavior in a regime when the number of measurements tends to infinity and the inter-sampling interval length is fixed. We present numerical simulations indicating that the exponential fitting algorithm ESPRIT is first-order optimal, namely, its first-order condition numbers have the same asymptotic behavior as the analytic ones in this regime.

ABSTRACT. The continuously increasing amount of noisy data demands the development of accurate and efficient models for analysis, modeling, and control. In this article, we propose a novel data-driven moment matching method which employs Tikhonov regularization in the Reproducing Kernel Hilbert Spaces (RKHSs). Specifically, considering a realistic scenario in which the system's plant is unknown and only noisy measured data are available, we provide an estimation of the moment of the unknown plant by solving a regularized optimization problem on RKHS. For, we first demonstrate that the estimation of the moment can be improved via tuning the regularization term, and further, we show under which condition the effect of the transient improves the performance of the estimation. Then, we construct a parameterized model characterized by a kernel-based output mapping. Finally, the proposed data-driven approach is validated and discussed by means of a DC-to-DC Ćuk converter driven by a Van der Pol oscillator.

ABSTRACT. In recent years, Industrial Cyber-Physical Systems (ICPS) within Industrial Control Systems (ICS) have faced increasing cyber-attacks due to greater connectivity, adoption of open-source protocols, and use of off-the-shelf products. These attacks can cause cyber anomalies like unauthorized IP addresses and data flow disruptions, potentially leading to physical asset breakdowns and compromising the ICS infrastructure. In this regard, it is crucial to ascertain whether these physical failures stem from cyber anomalies or from malfunctions in the physical components due to factors like aging, insufficient maintenance, or intentional physical sabotage [1,2]. Cyber-Physical Anomaly Detection (CPAD) algorithms address the joint monitoring of cyber and physical data that result in the detection of cyber, physical, and cyber-physical anomalies over the ICS. Such algorithms fall into three categories: signature-based, detecting known attacks via a database; behavior-based, identifying anomalies by learning normal behavior using data-driven approaches like machine learning; and specification-based, detecting anomalies by modelling ICS through probabilistic or mathematical methods. Recently, scientific literature has been investigating more and more the possibility of merging both behavior-based and specification-based algorithms to reduce false positives, commonly produced by behavior-based methods, and to lower the computational load associated with mathematical and probabilistic modelling, the primary drawback of specification-based approaches. Nevertheless, there is still little attention on deploying such “hybrid” algorithms to both the cyber and physical domains of the ICS, as the scientific literature mainly focuses on the monitoring of cyber anomalies [3]. Moreover, considering the distributed nature of ICPSs which are mainly deployed in critical infrastructures, like water treatment, and smart grids, there is a need to provide distributed multi-source and multi-modal CPAD algorithms that can detect anomalies in the different assets of the same infrastructure. Such algorithms may leverage data, feature, and decision fusion techniques that can merge the predictions of multiple, possibly redundant, detectors with the aim of making an agreed decision about an ongoing cyber, physical, or cyber-physical threat. Nevertheless, decision fusion techniques suffer from the possible presence of faulty detectors which could negatively weigh the final decision [4]. This motivates the dynamic self-adaptation of decision fusion to run-time uncertainty in order to prevent performance degradation and increase fault tolerance. Such self-adaptation is enabled by decision fusion’s explainability, which can be defined as the ability of decision fusion to provide insight into how many, which, and to what extent possibly faulty detectors contribute to the final decision.

Considering the limitations in current scientific literature, this work proposes two novel frameworks, namely “A hybrid behavior-and Bayesian network-based framework for cyber–physical anomaly detection” and “A two-level fusion framework for cyber-physical anomaly detection”.  

The first contribution proposes a novel hybrid “multi-formalism” CPAD framework for combining the outcomes of unsupervised behavior-based anomaly detectors applied to cyber and physical data through the adoption of a static Bayesian network. In more detail, the framework consists of two behavior-based anomaly detection modules that monitor separately and simultaneously the behavior of cyber and physical data acquired from the ICPS in order to detect and localize cyber, physical, and cyber-physical anomalies over the plant. By filtering and combining the outputs of the behavior-based anomaly detection modules through a Bayesian network-based modelling, the framework is able to: (I) provide the detection probability of cyber, physical, and cyber-physical anomalies; (II) improve the trustworthiness of the detected anomalies by taking into account the anomaly detectors’ confidence about their decisions; and (III) localize the detected anomalies, taking into account possible cascading effects over the cyber-physical process.

The second contribution proposes a novel flexible framework for multi-source and multi-modal CPAD which enables the combination of multiple and redundant CPAD detectors by means of a Time-Varying Dynamic Bayesian Network (TV-DBN) implemented as an explainable decision fusion technique. The framework involves two phases, namely the setup and inference phases. The former involves splitting the referenced ICPS into distributed subsystems and implementing the first detector-level fusion by training CPAD detectors on historical cyber-physical data, whereas the latter involves collecting evidence in terms of crisp labels from the multiple ICPS distributed subsystems and processing such evidence through the second ensemble-level fusion. Explainable ensemble-level fusion is implemented by a decision fusion technique that involves proportional voting inside a self-adaptive TV-DBN. This approach penalizes faulty detectors, excluding them from the decision fusion process if their reputation falls below an acceptable threshold. The framework and explainable decision fusion technique allow for: I) weighing evidence from CPAD detectors by taking into account the confidence about their decision; II) evaluating CPAD detectors’ behavior, i.e. their reputation, over time, based on their past decisions; and III) dynamically including or excluding CPAD detectors when performing ensemble-level fusion.

ABSTRACT. The problem of controlling a System of Systems (SoS) has been tackled with both distributed and centralized approaches. The distributed strategy is usually adopted when different agents may have a partial observation of the whole system and can control a subset of the subsystems by sharing information among them. The goal of this work is to provide a general control algorithm that can be used to control in a distributed manner an unconstrained multi agent System of Systems, while also testing it in a real environment where disturbances may happen. In particular, the coordination problem was formalized and solved by using a combination of the Model Predictive Control (MPC) paradigm, Linear Quadratic Tracking (LQT) techniques, and Consensus Alternating Direction Method of Multipliers (C-ADMM).

ABSTRACT. Over time, Single-Input Single-Output (SISO) systems have received significant attention in the field of control systems. However, real-world applications often involve Multi-Input Multi-Output (MIMO) systems, where the challenges associated with multivariable control are considerably greater. This abstract presents an innovative extension of the Set Membership Data-Driven (SMDD) approach from SISO to MIMO systems. Through a comparative analysis with the Virtual Reference Feedback Tuning (VRFT) method, it is demonstrated that the SMDD approach significantly outperforms VRFT, achieving notable reductions in decoupling errors. Furthermore, the SMDD controller exhibits faster rise and settling times, and completely eliminates overshoot. These findings indicate that the SMDD approach effectively enhances decoupling and error minimization, making it a more reliable solution for managing the complexities of MIMO systems.

ABSTRACT. Sensors fault is a critical issue for any kind of application. Abrupt sensor fault are frequent, but the most common situation shows sensors that slowly vary their sensing capabilities along time, generating the so-called Concept Drift (CD). CD handling techniques are based on the possibility to collect new supervised data and estimate the drift effect according to their time variation. The Water Quality Moni- toring (WQM) field is affected by the issue since electrochem- ical sensors degradation is common due to corrosion agents and the difficulties in periodically reaching each node of the WQM sensor network for proper maintenance. This paper discusses a possible approach for CD handling in WQM by using Machine Learning (ML) classifiers as “fake-supervisors” for labeling incoming new data. To this aim, the classifier must be reliable inside a certain neighborhood of the original training dataset centroids. Hence, by collecting data through a commercial WQM sensor, we simulate the aging process by means of an electrochemical corrosion model. Different types of corrosion trends are considered. The performances drop of three popular ML classifiers (namely Multy-Layer Perceptron Neural Networks, K-Nearest Neighbors, and Decision Trees) are compared when both raw data and data processed with PCA transformation are evaluated. Experimental results show that Neural Networks appear more robust to data perturbation, moreover the use of a PCA pre-processing step have a consid- erable positive effect in the model performance degradation trend.

ABSTRACT. The collective response to an emergency is a complex emergent phenomenon, whose study is of paramount importance in social science and risk management. However, few analytically-tractable mathematical models have been proposed in the literature to study such an important phenomenon. On the one hand, social-psychological efforts typically focus on elucidating the individual-level risk interpretation process, overlooking how this leads to collective emergent behaviors. On the other hand, classical mathematical models focus on an abstract representation of opinion dynamics, neglecting the specific characteristics of risk perception. To the best of our knowledge, few agent-based models have been proposed for this purpose, but they rely on complex mechanisms which hinder establishing rigorous analytical results, ultimately calling for the development of a novel analytically-tractable model for collective risk perception.

In particular, we fill in this gap by leveraging the mathematical theory of opinion dynamics. Specifically, we consider a scenario in which an institution broadcasts an estimate on the nature of the risk of a given emergency. However, it is known from the social psychology literature that individuals, besides receiving institutional information, they also exchange information with their peers and re-elaborate it on the basis of their own risk perception. Our model encapsulates these key features, ultimately yielding a realistic and analytically-tractable framework to study how social interactions and individual traits shape the collective risk perception.

ABSTRACT. We investigate an efficient and near-optimal method for variables identification of a standard PK-PD model for anesthetic drugs. Identified parameters act non-linearly in the construction of the formulation of the Hill function, which elaborates the output of the PK-PD model to furnish the bispectral index, a dimensionless number measuring the patient’s depth of hypnosis. The global input-output problem has the structure of a Wiener model, for which we provide a general Branch and Bound method for identification of non-linear variables whose efficacy is confirmed by numerical experiments.

ABSTRACT. Models of opinion formation have garnered significant attention across the interdisciplinary research community, captivating scholars and scientists ranging from sociologists to engineers and physicists alike. Bounded confidence models, perhaps the simplest nonlinear models of opinion formation, are distinguished by their rich and sophisticated dynamics. The classical Hegselmann-Krause model depicts the evolution of scalar opinions under “homophilous” social influence among social actors. In this model, each actor adjusts their opinion based solely on the similar opinions of their peers while rejecting dissimilar opinions. The prevailing generalizations to the multidimensional opinions case typically characterize the similarity of opinions based on their proximity in the Euclidean or some other norm. Instead, we consider a generalized model wherein two opinions are considered similar if their deviation falls within a specified confidence set. This set is not necessarily a ball and can be unbounded. We prove some convergence properties of the model at hand under mild assumptions on the confidence set; we also classify all equilibria of the model.

ABSTRACT. Dielectric Elastomer Actuators (DEAs) are highly attractive in the context of soft robotics due to their high flexibility, large deformations, and self-sensing capabilities. However, DEA-based soft robots face control challenges, such as nonlinearities and complex dynamics, complicating the design of model-based control strategies for position regulation and trajectory tracking. Additionally, the high voltage required to actuate DEAs often necessitates the use of bulky and expensive voltage amplifiers, which hinder portability and cost-effectiveness. This extended abstract briefly summarizes recent solutions proposed by the authors to address these challenges.

ABSTRACT. DC-DC power converters are pivotal components in modern electronics for voltage regulation. Standard PID controllers tend to dominate the manufacturers' production for industrial applications and replacing the low-level controller is often infeasible or unsuitable. In this work, we propose an adaptive reference governor to improve the transient behaviour of DC-DC converters, exploiting a linear time-variant Model Predictive Control based on a black-box model that is identified in real-time. It represents a plug-and-play solution that: i). fits a wide range of converters, ii). is self-tuning (i.e., no prior modeling effort), iii). is proven to run in real-time on cheap, conventional control boards, iv). does not require to alter the original controller. The solution is experimentally validated on a buck converter.

ABSTRACT. This extended abstract addresses the modeling and control of modular multilevel converters.

ABSTRACT. We model and control an impacting electromagnetic actuator using hybrid Lyapunov techniques. The proposed model accounts for the main magnetic effects and the mechanical stops inherent in the system’s design, which impose limitations on its range of motion. Additionally, the objective is to synthesize a control law with reduced complexity, enabling precise regulation of the position of the actuator’s moving part. The proposed control strategies include an LQ-based method for ensuring local stability and an energy-based backstepping approach for enhanced performance.

ABSTRACT. Pulse Width Modulation (PWM)-based multi-chamber Hydraulic Servo Actuators (HSAs) are highly efficient actuators for which it is possible a precise control of flow-rates and pressures. However, their modeling and control is not trivial due to the complexity introduced by the exponential increase in force combinations and their hybrid behaviour. This research work [1], on the one hand, extends the Average Equivalent Discrete-time Model (AEDM) of [2] to the mechanical dynamics, relating the valves duty-cycle with chamber pressures and piston force. On the other hand, it presents an adaptive control, combining a Decoupled Proportional-Integral (DPI) controller with a Model Reference Adaptive Control (MRAC) for an effective force tracking and pressure control, offering a better-performing alternative to existing methods.

References: [1] A. Bozza, G. Cavone, R. Carli, and M. Dotoli, “Pwm-based energy-efficient adaptive control for multi-chamber hydraulic servo actuators,” IEEE Transactions on Control Systems Technology, (Under submission).

[2] A. Bozza, G. Cavone, R. Carli, and M. Dotoli, “A power electronic converters-inspired approach for modeling pwm switched-based nonlinear hydraulic servo actuators,” in 2023 IEEE International Conference on Systems, Man, and Cybernetics (SMC). IEEE, 2023, pp. 2477–2482.

ABSTRACT. The electrification trend is spreading not only in the field of traditional vehicles, but also in the racing one, pushing towards the design of dedicated energy storage systems to cope with high demanding requirements, as witnessed by the next generation of Formula E that will increase power up to 600 kW. In this scenario, Hybrid Battery Packs (HBPs) raise up to an interesting solution thanks to their combined high-power and high-energy features. In this report, firstly, we showcase the potential benefits of hybrid battery packs, compared to single chemistry ones, when employed in a racing scenario. To this aim, we formulate a co-design optimization problem (Co-OP) to find the optimal hybrid battery pack configuration minimizing the race time. As a byproduct of the solution, the optimal power split between the two energy sources is obtained as an implicit control law. From this result, we address the problem of designing a real-time, explicit, energy management strategy (EMS) for the application. In this respect, we devise a modified co-design optimization problem, where explicit power split logic can be directly included, allowing to compare and rank them with respect to the optimal implicit one, considering as a case study the Formula E Gen 3 car in the 2021 Rome Formula E ePrix.

ABSTRACT. This work contributes to the railway field by proposing a novel energy management approach based on model predictive control (MPC) to provide a more efficient govern of the electric equipment inside trains. Specifically, the proposed MPC is capable of addressing the challenging scenario of partially catenary-free tracks for trains equipped with batteries. In particular, the MPC optimizes the current flows to supply the traction motors and the parallel-connected auxiliary loads, while reducing the amount of losses over the electric lines.

ABSTRACT. As the global focus on combating environmental pollution intensifies, the transition to sustainable energy sources, particularly in the form of electric vehicles (EVs), has become paramount. This paper addresses the pressing need for Smart Charging for EVs by developing a comprehensive mathematical model aimed at optimizing charging station management. The model aims to efficiently allocate the power from charging sockets to EVs, prioritizing cost minimization and avoiding energy waste. Computational simulations demonstrate the efficacy of the mathematical optimization model, which can unleash its full potential when the number of EVs at the charging station is high.

ABSTRACT. Vehicle-to-grid (V2G) technology has proven to be a promising solution for integrating electric vehicles (EVs) into the electricity grid, offering benefits such as grid stabilization and demand response. Predicting the aggregated available capacity (AAC) of EVs is crucial for the effective utilization of their energy storage capabilities. A comprehensive framework for predicting AAC in V2G systems is presented here. It includes methods for data pre-processing and feature selection tailored to managing complex datasets with multiple data sources such as GPS, weather, vehicle characteristics, historical data, and calendar information.

ABSTRACT. In multi-robot systems, relative localization is crucial in many tasks, such as leader following, target tracking, or cooperative maneuvering. Current approaches either rely on infrastructure-based or on infrastructure-less setups. The former typically achieve high localization accuracy but require fixed external structures. The latter provide more flexibility, but often requires Line-of-Sight (LoS) devices like cameras or lidars. Ultra Wide Band (UWB) devices are emerging as a viable alternative to build infrastructure-less solutions that do not require LoS. These approaches directly deploy the UWB sensors on the robots. However, they require that at least one of the platforms is static, limiting the advantages of an infrastructure-less setup. In this work, we remove this constraint and introduce an active method for infrastructure-less relative localization. Our approach allows the robot to adapt its position to minimize the relative localization error of the other platform.

ABSTRACT. This paper proposes a framework for controlling a drone swarm to achieve two goals: i) covering a desired region of interest through onboard cameras that capture videos to be sent in real-time to a Ground Control Station (GCS), and ii) ensuring the highest video quality possible given the available Internet network bandwidth. Indeed, the quality of received videos depends on both the available bandwidth and the altitude of drones, which influences pixel density. Thus, contrary to the conventional assumption of uniform drone altitudes, we let drones to track a reference altitude that is function of the time-varying available bandwidth to improve visual quality. To achieve the aforementioned goals, we propose a leader-follower multi-agent system formation control problem. In particular, the leader tracks a desired path using Nonlinear Model Predictive Control (NMPC) to cover the area of interest. Follower agents track the leader using NMPC, aiming at maximizing both the total coverage area and the quality of the videos sent to the GCS, considering the constraints imposed by the network available bandwidth. At the same time, we formulate the NMPC problem to ensure that the swarm maintains a formation characterized by a given overlap percentage between the videos captured by the drones while avoiding collisions. This allows dynamical stitching of the received videos at the GCS, enabling the execution of computer vision algorithms for tasks such as object detection and surveillance.

ABSTRACT. Redundant robots have the potential to perform internal joints motion without modifying the pose of the end-effector by exploiting the null-space of the Jacobian matrix. Capitalizing on that feature, we developed a control technique for minimizing the robot visual appearance when observed from an eye-to-hand camera. Such algorithm is instrumental in contexts where quickly adjusting the perspective to see objects obstructed by the robot is impractical (e.g., teleoperation in narrow environment). Diminished reality techniques are frequently employed in these cases to mitigate the robot intrusion into the environment, although these techniques may sometimes compromise the perceived realism. The experimental evaluation confirmed the effectiveness of our control algorithm, demonstrating an average reduction of 4.67% of the area covered by the robot within the frame when compared to the case without the optimization action.

ABSTRACT. Classical position-based and image-based visual servoing methods suffer from some drawbacks, the former require a three-dimensional model of the observed scene and they are sensitive to camera calibration errors, the convergence of the latter cannot be ensured in the whole feature space. In the last two decades, the homography-based visual servoing approaches have been widely investigated. They overcome the above drawbacks and avoid the intermediate process of feature extraction. The main results on homography-based control schemes are formulated in the continuous time domain. Due to the limited camera frame rate, actuation delay, and computation time, the intrinsic sampled-data nature of such control schemes should be taken into account. To this aim, in this paper, we propose an exact sampled-data model of the homography dynamics and an adaptive visual controller that directly uses the homography matrix and explicitly takes into account camera velocity limits ensuring global asymptotic stability at the same time.

ABSTRACT. Autonomous robot navigation relies on its ability to understand its environment for localization, typically using a Visual Simultaneous Localization And Mapping (V-SLAM) algorithm that processes image sequences. In this work, we propose LF2SLAM, “Learned Features For SLAM”, a hybrid method for navigation in challenging indoor scenarios, like low-light conditions. It integrates a robust data-driven feature extraction within a classic geometric VO estimation pipeline. In particular, we train a deep neural network for feature extraction and integrate it into one of the most used state- of-the-art algorithms: ORBSLAM3. We developed a novel loss function employing a binary mask for filtering the informative features. The experimental evaluation shows that our approach has remarkable generalization capabilities in scenarios that differ from those used for training.

ABSTRACT. This study presents a real-time hand pose extraction method through computer vision, utilizing a single RGB camera to facilitate human-computer interaction. The implementation, designed for current consumer-level computing power, employs a probabilistic skin color distribution derived from the user’s face detected by a Haar classifier. The system’s effectiveness is demonstrated by creating and manipulating sound in real- time interaction. While robust, the method assumes a single individual with long sleeves and avoids skin-like colors in the input image. The proposed approach combines face detection, skin color distribution, and blob labeling with a careful thresh- old adjustment and Histogram of Oriented Gradients (HOG) features, showcasing its potential for intuitive and non-intrusive human-computer interaction.

ABSTRACT. This extended Abstract reports recent developments, by the author, in the analysis of dynamical systems and educational activities. Analysis. Starting from the recent findings regarding self-similarity in walking and swimming (in its simple and enhanced nature), a new research direction - involving the tennis framework and in particular relying on experimental results on the forehand - has been defined and highlighted, with the theoretical derivations and the experimental results being in line with all the previous research carried out within different frameworks. Education. "Kids in Control" is a control-centered workshop that has been conceived for promoting STEM (Science, Technology, Engineering, and Mathematics) skills and inclusion for kids from 8 to 10 years old. It includes activities and devices developed to improve the learning process, while fostering the creation of a new teaching and learning style, based on the most advanced digital technologies. It also aims at being firmly rooted in the foundations of pedagogy and in the interaction between teacher and learner.

ABSTRACT. This contribution presents a structured approach to teaching linear system identification at the undergraduate level using blended learning (BL) techniques and advanced assessment strategies. Developed at Università Politecnica delle Marche, the course integrates both in-person and online educational components, leveraging tools like MathWorks MATLAB and Moodle LMS to enhance learning outcomes.

The aim of this contribution is twofold: firstly, to provide a tangible example of implementing the BL approach in university courses; and secondly, to introduce current trends and discussions on the potential impact of new technologies and teaching strategies in meeting the evolving needs of control engineering education.

The attached manuscript highlights the development of the course's online infrastructure, including short video lessons, MATLAB Academy's online courses, and the MATLAB Grader tool for formative assessments. These resources not only reinforce students' understanding of key concepts but also improve their proficiency in MATLAB programming and related skills.

The implementation of BL in this course yielded positive results, such as enhanced student engagement and improved learning outcomes. However, there are areas for improvement, including better balancing of physical and online components and addressing challenges related to the allocation of time for online resources. Future iterations aim to refine these aspects to further optimize the learning experience.

ABSTRACT. The evolution of teaching methodologies in automatic control has revealed significant challenges and opportunities for collaboration among educators. Two main issues that influence workflows are: 1) the systematic difficulty in exchanging instructional materials with colleagues, and 2) the complexity of establishing commonalities in the courses’ programs across different institutions and nations. These challenges are interconnected, impacting both the efficiency of teaching and the enrichment of student experiences. IFAC’s Technical Committee (TC) 9.4 is investigating solutions to these problems that rely on open databases of instructional materials as enhancers of teaching and learning. Such databases offer valuable content for teaching activities, scaffolding, and self-assessment provided that robust taxonomies are identified and used to index the instructional materials to avoid the subjective biases. The attached contribution provides an overview of the solution being investigated by this TC and how it may improve collaborative teaching and learning in automatic control.

ABSTRACT. Drivers' conduct is heavily impacted by their perception of risk, which can compromise road safety and efficiency and have a major effect on the transportation environment. This work tackles this important problem by presenting an algorithm that uses electroencephalogram (EEG) data to estimate driver-perceived danger. The method builds a strong prediction model for predicting perceived danger levels using a Support Vector Machine (SVM), which may subsequently guide the deployment of preventative safety measures. In the simulation environment, SCANeR Studio, players had to make choices as they entered and navigated a two-lane roundabout full of moving cars.

ABSTRACT. Guaranteeing data confidentiality while allowing agents to reach an agreement on some shared variables is an essential feature to foster the adoption of distributed protocols. In this submission, we take a geometrical perspective and we implement a novel encryption methodology via coordinate shifts. This choice guarantees, on one side, the existence of an uncountable infinity of possible vectors and, on the other side, provides a control-theoretical viewpoint on encryption that naturally blends with a dynamical system. Utilizing vector coordinate shifts, our method excels in efficiency and speed of implementation, ensuring compliance with the stringent time constraints of real-time protocols. This makes our method easily applicable to a wide range of critical infrastructures, including the GOOSE protocol for communication of digital power substations. Indeed, adherence to the IEC 62351 standard governing security requirements for the GOOSE protocol has always been an open challenge as it presents practical problems due to the conflict between time requirements imposed by operations and traditional security procedures.

ABSTRACT. This contribution presents the project MUSAPOEM (Multi Satellite Proximity Operations for Rendezvous and Docking Missions in Earth and Moon Orbits). The Italian Space Agency co-funds the project and sees the collaboration between two universities and companies: the Polytechnic of Turin, the University of Pisa, Argotec and Kurs Orbital. Its objective is to advance the state-of-the-art in modelling, guidance, navigation, control, and identification algorithms for multi-vehicle autonomous proximity operations in space, defined as operations where the relative distance among the satellites is less than a few tens of kilometres, with a particular focus on the vicinity of the Moon.