ABSTRACT. The stability analysis of linear time-varying (LTV) systems remains a fundamental challenge in dynamical systems theory, as the instantaneous eigenvalues of the state matrix often fail to provide necessary or sufficient conditions for stability. This work introduces a novel framework for the stability analysis of LTV systems by defining a spectral decomposition based on a specific linear operator. By solving the associated operator eigenproblem, we show that once a set of pointwise linearly independent eigenfunctions is obtained, the state transition matrix admits an explicit, closed‑form representation. Furthermore, we derive new sufficient conditions for asymptotic stability accounting for the interplay between the dynamical eigenvalues and the time-variation of the coordinate transformation. The framework is validated using Rugh's Example 8.1, i.e., a well-known benchmark from the literature.

ABSTRACT. A new method for constructing asymptotic Bode plots is proposed, using generalized approximating functions. The proposed method is referred to as a direct method since it allows to directly draw the asymptotic Bode magnitude and phase plots of the complete transfer function without requiring the detailed analysis nor the plots construction of each factor.

ABSTRACT. The invariance principle, through which the steady-state behavior of nonlinear systems was introduced by Isidori and Byrnes, is leveraged in this article to bring forth a characterization of the frequency response of nonlinear systems. We show that, for systems under nonlinear periodic excitations, the frequency response can still be defined as a complex-valued function in a phasor form. However, together with suitable notions of gain and phase functions, we show the existence of another function that completes the frequency response and allows quantifying the distortion introduced by the system in the steady-state output.

ABSTRACT. This work addresses the problem of robust output regulation for single-input single-output bilinear systems subject to input saturation, in the presence of constant references and disturbances. Two output-feedback control strategies based on integral action are proposed. The first combines a forwarding-based state-feedback design with a Luenberger-type observer through a separation-principle framework, while the second relies on a pure integral feedback law. Both strategies are formally proven to achieve robust regulation. The first strategy is experimentally validated on a counter-current heat exchanger, where the outlet temperature is regulated by manipulating the flow rate. Experimental results demonstrate significant advantages over standard proportional-integral control, including improved tracking, saturation avoidance, and reduced resource consumption.

ABSTRACT. Finite- and infinite-horizon optimal control problems are deeply connected, yet they are often studied through distinct theoretical frameworks due to their different analytical and computational challenges. In this work, we investigate this connection through a notion of ``pattern-preservation'', which characterizes the persistence of structural features of finite-horizon optimal controls in the infinite-horizon limit. Building on previous results based on $\Gamma$-convergence, we show that dissipativity of the control systems (as introduced by Willems) combined with a coercive storage function provides a verifiable sufficient condition for such property in a broad class of optimal control problems. This establishes a formal link between dissipativity and pattern preservation, showing that dissipative systems naturally preserve optimal control structures across finite and infinite horizons.

ABSTRACT. The stability properties of gated Recurrent Neural Networks (gRNNs) have received significant attention in recent years. This paper analyzes the Input-to-State Stability and Incremental ISS (δISS) of the relatively unexplored Chaos-Free Networks (CFNs), proving their intrinsic ISS by design. Building on this, we propose a novel variant, the Decoupled-Gate Network (DGN), and show that it is unconditionally ISS and δISS. Compared to state-of-the-art ISS and δISS gRNNs, which require training modifications to enforce stability, DGNs preserve standard training pipelines. Numerical results show that stability is guaranteed without a significant deterioration in accuracy.

ABSTRACT. Leaky Integrate-and-Fire (LIF) neurons are widely used to model neuronal firing because they provide a simple yet biologically meaningful description of membrane potential dynamics. We analyzed the stability of two diffusively coupled LIF neurons with threshold-reset dynamics. The system forms a hybrid dynamical model in which continuous voltage evolution is interrupted by instantaneous resets whenever a neuron reaches its firing threshold. We reduced the hybrid model to a one-dimensional spike to-spike Poincar´e map, making the stability analysis tractable. We proved the existence of two limit cycles: an unstable synchronous oscillation and a stable phase-locked oscillation. For the non-leaky case (µ = 0), a closed-form expression of the Poincar´e map is derived using the Lambert W function, enabling complete stability analysis. For the leaky case (µ > 0), almost global exponential stability is established for sufficiently large coupling strengths. The work provides theoretical insight into synchronization and phase-locking in coupled hybrid neuronal systems.

ABSTRACT. In this paper, we introduce a minimal dynamical model for the diffusion of a new social norm, in which individuals transition among three states: non-supporters, silent supporters, and vocal advocates. Despite its simplicity and its close relation to the classical SI-type and SIS-type spreading dynamics, this model exhibits a nontrivial latent–outbreak dynamic pattern: an initial small adoption wave is followed by a long quiescent period and then an abrupt, endogenous explosion of support.

ABSTRACT. This extended abstract shows a new formulation for the dynamic model of nonlinear mechanical systems, that can be applied to different automotive and robotic case studies, and a modeling procedure allowing to automatically obtain the model formulation.

ABSTRACT. This paper presents a unified control strategy for robotic manipulators that ensures safe manipulator motion by simultaneously addressing non-convex obstacle and singularity avoidance within a single optimization framework. The proposed approach leverages Control Barrier Functions (CBFs) to define two complementary safety constraints: one formulated in the Cartesian space to maintain a safe distance between the end-effector and surrounding obstacles, and another defined in the joint space to prevent the manipulator from entering uncontrollable configurations. Non-convex obstacle avoidance is achieved through an efficient sampling-based CBF approach. Both CBFs are integrated into a Quadratic Programming (QP) structure that computes a minimally invasive control input. The method was validated in simulation, demonstrating effective avoidance of both collisions and singular configurations without compromising convergence to the desired target.

ABSTRACT. Selective harvesting of fresh-market crops is labor-intensive because fruit must be picked individually and gently in cluttered, unstructured environments. This paper addresses the challenges of motion planning and force-controlled manipulation for robotic fruit harvesting through the development of a tomato-picking robotic system. The proposed platform employs a collaborative manipulator to harvest tomatoes with varying ripeness, size, and color, often arranged in clusters or partially hidden by leaves and branches. Beyond fruit detection and localization, safely approaching the target and detaching it without causing damage represent critical challenges. To this end, the system adopts a compliant soft gripper that provides intrinsically gentle and non-damaging grasping. The first contribution of this work is a multi-stage planning strategy that coordinates scene observation, collision-aware approach, and grasp execution in complex and cluttered environments. The second contribution is a force-controlled manipulation strategy that continuously monitors the interaction forces during grasping and detachment. Since the selected soft gripper does not incorporate an integrated cutting mechanism, fruit separation relies on a pulling-based detachment action, making force monitoring essential to avoid excessive loads that could damage either the fruit or the plant while ensuring reliable harvesting.

ABSTRACT. Human-inspired motion generation plays a key role in enabling robotic manipulators to operate safely and naturally in environments shared with humans. This paper presents an overview of our recent research activities on human-inspired motion planning for robotic manipulation. Starting from the functional analysis of human upper-limb movements, we developed a computationally efficient planning framework that embeds human kinematic characteristics into robot motion generation. The proposed methodology has been progressively extended to address increasingly challenging manipulation tasks, including dual-arm handling of large objects, dynamic grasping of moving targets, and safety-aware motion generation during human-robot interaction. Experimental validation across these scenarios demonstrates the versatility of the proposed framework and its ability to generate smooth, human-like motions while maintaining high task performance.

ABSTRACT. This paper presents a hierarchical Hidden Markov Model (HHMM) for programming robotic manipulation skills from intuitive human demonstrations. The framework models manipulation primitives while explicitly encoding grip force within the learned skill representation. Human demonstrations are acquired through sEMG-based inputs and tactile sensing, enabling robust inference despite signal variability. Experiments on a collaborative robot with a sensorized gripper demonstrate reliable reproduction of motion and grip-force profiles, highlighting the effectiveness of the proposed approach for intuitive programming by demonstration.

ABSTRACT. Human-robot collaborative logistics require scheduling strategies capable of adapting to the stochastic behaviour of human operators while preserving safety and production efficiency. This work presents a reinforcement learning framework for dynamic task scheduling in collaborative workstations based on a high-fidelity digital twin developed in Siemens Tecnomatix Process Simulate. A Maskable Proximal Policy Optimization agent learns scheduling policies from interaction with the simulated environment while action masking guarantees compliance with production constraints and safety requirements. Human behaviour is represented through a context-dependent finite-state machine reproducing realistic operator variability. Experimental results demonstrate reliable convergence and significant improvements over random scheduling strategies, achieving complete task execution while maintaining safe human-robot collaboration.

ABSTRACT. This extended abstract adopts a strategy inspired by the micro/macro manipulation paradigm to control mobile manipulator in pHRI, which explicitly uses the complementary properties of the two subsystems: the virtually unbounded workspace and slower dynamics of the mobile base, and the limited workspace but faster dynamics of the lightweight manipulator.

ABSTRACT. The purpose of this work is the integration and assessment of modern generative Artificial Intelligence text chatbots in combination with social robots, which permits to establish a human-friendly and human-like interaction. In detail, a human-robot vocal interaction scheme is presented, using OpenAI text generation service application programming interface in combination with a robot. The scheme implementation consists of delegating the messages management to an external device linked to the robot, which then provides feedback returned to the user via the robot. This approach was technically evaluated in terms of real-time use and computational burdens. Finally, the robotic platform was tested with a group of 23 users, who were asked to interact with the robot and fill in a survey to assess the quality of human-robot vocal interaction.

ABSTRACT. A Pontryagin-based receding-horizon framework is proposed for autonomous human-drone rendezvous in industrial scenarios. The method combines a free-final-time optimal control formulation with an indirect Boundary Value Problem solved online to continuously adapt the rendezvous trajectory. Safety, ergonomic feasibility, and energy efficiency are integrated into a unified optimization framework. Numerical simulations demonstrate accurate, adaptive, and energy-efficient rendezvous under time-varying human motion.

ABSTRACT. Robot collision detection and isolation is successfully achieved without the need of external sensing by the well-known momentum-based residual method, typically implemented using the Euler-Lagrange formalism. Exploiting available recursive Newton-Euler methods for inverse dynamics and for computing the first derivative of the torque, a general-purpose O(n) algorithm is presented for the real-time numerical evaluation of the residual vector of any serial robot manipulator with n joints.

ABSTRACT. This paper presents a novel discrete-event (DE) formulation for the optimal allocation and scheduling of electric buses (EBs) charging operations within a parking lot. The proposed model is formulated as a Mixed-Integer Linear Programming (MILP) problem, allowing for the joint optimization of charging sequences and grid power usage under operational and temporal constraints. The resulting model is suitable for efficient computation using well-established techniques and algorithms. Compared to traditional time-indexed formulations, the proposed model achieves a significant reduction in the number of decision variables, by approximately one order of magnitude, while maintaining accuracy. The inclusion of grid capacity constraints leads to longer completion times compared to the unconstrained case, but provides improved feasibility with respect to grid power limits. The observed computational scalability, tested on a real case study provided by Iveco S.p.a., confirms the practical applicability of the model for mediumto-large EB fleets.

ABSTRACT. This work addresses autonomous vehicle carpooling under time-varying traffic and passenger-service requirements. Pick-up/drop-off sequencing is formulated as a Constrained Markov Decision Process (CMDP), while a traffic-aware routing layer converts each selected target into a feasible path. Time-window, waiting-time, ride-time, detour and dwell-time violations are encoded as causal soft costs and regulated by projected Lagrange multipliers. A masked PPO policy preserves pick-upbefore- drop-off causality.

ABSTRACT. This paper presents a distributed energymanagement architecture for scheduling electric-vehicle (EV) charging in smart districts. A day-ahead Integer Linear Programming (ILP) model minimizes charging costs while satisfying charging-point and district energy limits, distinguishing interruptible and non-interruptible sessions, and accounting for State-of-Charge (SoC) priorities and preferred starting times. To address large instances, the centralized problem is decomposed into local ILPs solved in parallel by a distributed matheuristic. A Kubernetes cluster hosts the master controller and solver agents, whereas a permissioned Hyperledger Fabric blockchain provides reliable, privacy-aware, and tamperresistant data exchange. A case study with 10 Charge Point Operators (CPOs), 20 charging points, and 40 EVs shows that the proposed method preserves grid feasibility, improves userpreference satisfaction, and achieves better objective values and computation times than a Genetic Algorithm benchmark.

ABSTRACT. Proactive Quality of Service (QoS) estimation is crucial for reliable, low-latency 6G vehicular networks. Leveraging an OMNeT++/Simu5G and Simulation of Urban Mobility co-simulation, a high-fidelity dataset of urban spatiotemporal network dynamics is generated. A Random Forest regressor is trained using only two-dimensional vehicular geographical coordinates to predict a spatially aggregated QoS score, explicitly integrating normalized Signal-to-Interference-plus- Noise Ratio, Throughput, Medium Access Control delay, and Packet Loss Rate. Experimental evaluation compares Random Forest against standard regression baselines, showing that the model provides a competitive trade-off between prediction accuracy, non-linear spatial modeling capability, and computational efficiency.

ABSTRACT. Cold-chain transportation is affected not only by traffic congestion but also by the environmental conditions experienced by refrigerated vehicles during travel. Conges- tion increases travel and labor costs while extending the period during which refrigeration must operate; at the same time, different routes and departure times may expose the vehicle to different temperature profiles. This contribution presents a forecast-driven online routing framework for con- nected refrigerated vehicles. Traffic-dependent travel times and environment-driven refrigeration costs are jointly incorporated into a time-dependent cold-chain routing problem. A path- based multigraph and a departure-time bucket discretization are used to obtain a tractable mixed-integer linear program- ming formulation. The problem is repeatedly solved within a shrinking-horizon architecture as updated traffic and weather forecasts become available. A realistic regional case study shows that environment-aware replanning can modify both route alternatives and customer visiting order, reducing refrigeration expenditure and total operating costs with respect to traffic-only and open-loop strategies.

ABSTRACT. Most of the research on transport planning has focused on optimizing revenues or utilitarian metrics such as average travel times, which often ends up penalizing the worst-off for the sake of efficiency. At the same time, most of the research in transport justice has focused on assessing injustices, without being able to prescribe operational solutions. This paper contributes to bridging this gap and presents optimization models for justice-informed operational planning of intermodal mobility systems that explicitly account for the budget and safety limitations of users, and for infrastructural capacity constraints. Specifically, we first focus on an intermodal Autonomous Mobility-on-Demand (AMoD) system—where self-driving robotaxis provide mobility jointly with public transit and active modes—and characterize its operations from a mesoscopic planning perspective via network flow models. Second, we leverage these models to optimize system operations through both utilitarian efficiency and justice-informed objectives. We showcase our framework in a real-world case-study for Manhattan, New York. Our results show that monetary budgets significantly limit the social justice of AMoD systems. At the same time, granting free public transit can result in sufficiency levels very close to a completely free intermodal AMoD system, where justice informed operations can be achieved without compromising standard efficiency metrics.

ABSTRACT. This work presents an event-triggered planning algorithm for the scheduling of last-mile delivery and battery charging operations of a heterogeneous fleet of autonomous mobile robots employed in pedestrian areas. Parcel clustering, route optimization, robot-route assignment and energy-aware charging decisions are all combined in a unified planning architecture. The main objectives of the proposed algorithm are the reduction of customer waiting times, the reduction of delivery travel times, and the maximization of energy exploitation. The results obtained on the Savona University Campus case study confirm the effectiveness of the proposed approach in enhancing delivery efficiency, service reliability and energy-aware charging management.

ABSTRACT. This paper presents the real-world experimental validation of GoA-based Autonomous Driving Function (ADF) control architecture for High-Speed Train (HST) over ETCS on the Italian San Donato test circuit. The leader of the manuscript is TINO, i.e., the first italian fully autonomous HST prototype provided by Rete Ferroviaria Italia S.p.A.. The experimental campaign shows that TINO, embedded with ADF, is able to guarantee the main ATO functionalities, i.e., the optimal driving profile generation and speed tracking. The results prove that ADF effectively enables the GoA4 for HST, a key pivot for the new virtual coupling railway mobility paradigm.

ABSTRACT. This paper develops a distributed state estimator for CACC systems affected by sensor faults and limited measurements. Unlike existing approaches, it explicitly accounts for time-varying communication delays while requiring the solely jointly observability assumption. Delay-dependent LMI conditions guaranteeing convergence are derived, whereas simulation results validate the theoretical findings.

ABSTRACT. Optimal control frameworks in Human-Robot Collaboration (HRC) frequently rely on static cost function weights, limiting the adaptability of robotic systems across varying safety conditions. This extended abstract presents a safety-aware dynamic weight adaptation rule for Quadratic Programming (QP) controllers. By utilizing a Control Barrier Function (CBF) derived from Speed and Separation Monitoring (SSM) guidelines as the environmental state variable, the framework continuously modulates controller priorities in real-time. The weight transition functions are mathematically formalized, and strict algebraic constraints are established to guarantee the positive definiteness of the cost function. Simulations demonstrate that while static parameters suffer from rapid performance degradation outside their designated tuning zones, the dynamic adaptation rule exhibits robust generalization, averting system freezing and preserving high on-target tracking rates across all safety transitions.

ABSTRACT. Continuous-time multi-agent path finding on roadmaps (MAPFR) provides a principled model for auto- mated guided vehicle (AGV) fleet coordination, but indus- trial roadmaps make repeated safe-interval construction and multi-agent replanning expensive. This paper introduces Lazy Continuous-Time Priority-Based Search (Lazy-CPBS), a cen- tralized solver that systematically explores partial priority orderings. Industrial roadmaps are represented in configuration space and enriched with offline continuous-time conflict anno- tations. When a new precedence relation is introduced, only the lower-priority agent and its descendants can acquire additional constraints. Lazy-CPBS postpones node materialization until expansion and replans only these affected agents. Experiments on standard benchmarks and five industrial roadmaps show that Lazy-CPBS achieves the strongest overall robustness– runtime compromise among the evaluated solvers.

ABSTRACT. This work presents a Quantum Model Predictive Control (QMPC) framework for nonlinear constrained optimal control and demonstrates its application to autonomous spacecraft rendezvous during proximity operations. The proposed methodology reformulates a nonlinear Model Predictive Control (NMPC) problem as a Quadratic Unconstrained Binary Optimization (QUBO) problem, enabling its solution through quantum annealers and quantum-inspired optimization algorithms. The approach combines three main ingredients: a move-blocking strategy for reducing the optimization dimension, a polynomial prediction model obtained through Maclaurin expansion, and a binary encoding of the control variables. A novel dynamic penalty mechanism is introduced to efficiently handle nonlinear state constraints without increasing the QUBO dimensionality through auxiliary variables. The proposed controller is validated on a realistic spacecraft rendezvous scenario involving Keep-Out Zone (KOZ) and Line-Of-Sight (LOS) constraints. Simulation results demonstrate that the controller generates safe rendezvous trajectories while maintaining favorable computational scalability, highlighting the potential of quantum optimization for nonlinear predictive guidance and control problems.

ABSTRACT. This study investigates metastable and hysteretic non-linear dynamics in real-time psychological data from Ecological Momentary Assessments (EMA). While standard approaches often overlook multi-state dynamics and path-dependency, we introduce a novel analytical framework based on Recurrence Quantification Analysis. To detect metastability, we utilize a Markovian optimization problem, validating the identified metastates against a dynamical graph structure of psychopathological networks. Results show that the number of metastates significantly decreases during transitions into depressive states. Furthermore, to capture hysteresis—the psychological system's memory—we propose a novel lag-detection pipeline that addresses delayed effects through temporal re-alignment, enabling the exact reconstruction of bifurcation diagrams. Validated on both real-world and synthetic datasets, this dual approach effectively maps complex driver-response interactions, offering new mathematical tools to anticipate abrupt systemic shifts in clinical psychology.

ABSTRACT. In this paper, we propose a preliminary study for the design of an adaptive data-driven control law for solving the model-reference problem of an unknown switched system. The proposed strategy aims to regulate the switched system such that the closed-loop behavior matches the desired dynamics defined by a reference model. By treating the switching signal as an external disturbance with sufficiently long dwell times between consecutive switching instants, we develop a model-reference design procedure that provides stability guarantees. Finally, simulation results demonstrate the effectiveness of the proposed approach.

ABSTRACT. Autonomous Unmanned Aerial Vehicle (UAV) flight in GPS-denied environments requires a perception stack that supports two geometrically related but traditionally disjoint tasks: avoiding obstacles and selecting landing sites. This paper proposes a unified architecture that applies a real-time, CPU-only macro-to-micro planar segmentation pipeline to a simulated quadrotor's onboard perception. Dense 3D point clouds are abstracted into Oriented Bounding Boxes (OBBs) and registered into a Simultaneous Localization and Mapping (SLAM) frame, creating a single, persistent geometric layer. Vertical instances are extruded as obstacles, while a lightweight heuristic filters horizontal instances to select the landing candidate closest to the mission end-target. Simulation results demonstrate the feasibility of this shared-layer approach for real-time edge computing.

ABSTRACT. The expansion of distributed energy systems increases harmonic distortion in the grid, driving stricter power quality regulations and the development of advanced control techniques for grid-connected inverters. This paper proposes repetitive current control based on internal model compensation (CIM) for grid-following inverters, addressing stabiliser conservativity and limited frequency adaptation capability of conventional repetitive control.

ABSTRACT. We present a stochastic control framework for electric-vehicle (EV) charging stations equipped with on-site photovoltaic (PV) generation and battery storage. Vehicles are aggregated into a finite number of classes according to their residual charging demand, yielding a compact state whose dimension is independent of the fleet size. On the corresponding expectation model we formulate a finite-horizon smart-charging problem, in the form of a linear program (LP), that jointly optimizes class-wise charging actions and energy-management variables, balancing electricity-purchase cost against customer satisfaction. The LP is solved in shrinking-horizon form and implemented online after integer discretization of the first action; a robust counterpart preserves feasibility under interval ambiguity on arrivals, departures and PV generation. The controller cuts the cost-per-kWh by up to 17.5% and the total daily cost by 10% to nearly 30% versus a service-greedy First-In-First-Served (FIFS) baseline, at only a marginal service-quality loss.

ABSTRACT. Accurate open-circuit-voltage (OCV) characterization is important for model-based battery management, since errors in the OCV–state-of-charge (SOC) relation affect voltage prediction and subsequent state estimation. This paper investigates Bayesian particle-based identification of a polynomial battery OCV–SOC model under the assumption that SOC is known. The uncertain coefficients are represented by particles, and voltage information is incorporated through a Gaussian likelihood and Bayes’ rule. Two posterior representations are compared: fixed particles with evolving weights and importance resampling with equal-weight particles. The method is assessed using laboratory characterization of a LiFePO4 battery controlled through a programmable two-quadrant test bench. Both formulations reconstruct the reference relation obtained from experimental data with small voltage residuals. For the reported settings, the fixed-particle formulation is slightly closer to the reference and computationally less demanding, whereas resampling produces a more concentrated particle population. The study provides a basis for future joint SOC–parameter estimation.

ABSTRACT. The increasing use of distributed renewable energy sources is changing how power systems are managed. In this context, Citizen Energy Communities are becoming important for improving local energy use and supporting the grid. This paper proposes a method to manage multiple communities in energy markets. A new entity, the energy community aggregator, is introduced to coordinate multiple communities and enable their participation in balancing and peer-to-peer markets. A bilevel optimisation model is developed: the upper level defines the aggregator strategy, which aims to provide demand response and peer-to-peer market coordination, while the lower level optimises energy costs for each community manager and its participants. The lower-level problem is reformulated using Karush–Kuhn–Tucker conditions. The system model includes various energy technologies, such as photovoltaic systems, energy storage, electric vehicles operating in vehicle-to-grid mode, microturbines, and hydrogen systems. Numerical results from a synthetic case study of three communities, involving 45 participants, show that the proposed approach improves coordination among communities, increases self-consumption, and supports grid services, while maintaining optimal economic benefits for users and communities. Moreover, analysis on sensitivity on various demand response signals and scalability on instances of increased sizes, up to 1125 users, shows the robustness of the proposed approach.

ABSTRACT. This work presented a distributed energy-management framework for multi-zone buildings with PV, BESS, and HVAC systems. A high-order RC thermal model is combined with an exact LP relaxation of a standard MILP formulation and a DC-ADMM distributed solver.

Theoretical analysis and numerical evidence indicate that the LP reproduces MILP optimality under mild assumptions, while the distributed implementation converges to the centralized solution with localized computations and communication.

ABSTRACT. Airborne Wind Energy Systems (AWES) are tethered aerial vehicles that harvest wind energy by flying at high speed in crosswind conditions. Reaching full autonomous and high-performance flight is a key milestone to improve technology readiness. This work addresses the periodic power oscillations inherent to circular crosswind flight of rigid-wing fly-gen AWES (windplanes) with onboard turbines, proposing a power-smoothing scheme based on cyclic modulation of the turbine tip-speed ratio within a flight architecture that confines the flight to a prescribed region of space without tracking a specific path. Closed-loop simulations against stationary and 8-hour time-varying wind fields demonstrate power oscillations reduced to within 5% of the mean, with only a 9% penalty relative to the open-loop power-optimal trajectory.

ABSTRACT. This work presents an energy-aware multi-agent mechanism for adapting edge-deployed machine learning models under changing operating conditions. Edge Agents monitor prediction error and select among local, peer-assisted, and cloud-assisted responses according to expected recovery, energy, and latency. Experiments on real IoT sensor streams show that cooperation remains close to systematic local retraining in accuracy while reducing adaptation energy

ABSTRACT. Climate-policy effectiveness depends not only on institutional design but also on how social inequalities, peer influence, and political feedback shape public acceptance. We develop an agent-based model of Italy based on a two-layer multiplex network connecting 100,000 citizens distributed across the 20 Italian regions with 895 regional political representatives. Citizens differ in income, gender, geographical location, green propensity, and self-efficacy, while their social connections reflect geographical, socio-economic, and gender-based interaction patterns. Green propensity evolves through peer influence and feedback from national or regionally targeted climate policies. We compare six policy designs under different assumptions regarding self-efficacy, peer influence, and gender segregation. Progressive policies generate the highest levels of public support and the most homogeneous territorial outcomes, whereas regressive policies produce lower acceptance and stronger North–South polarization. Peer influence reduces disparities across income groups, while lower self-efficacy among disadvantaged citizens decreases overall policy support. Gender segregation amplifies differences between men and women when their self-efficacy levels differ, whereas more integrated networks mitigate these gaps. Finally, policies targeting northern regions produce the largest increase in national support because of their population size and stronger network connectivity.

ABSTRACT. We propose a control-theoretic framework for constrained optimization with nonsmooth composite problems. Using the proximal augmented Lagrangian, we define a fictitious plant whose equilibria correspond to the problem's stationary points. Treating the Lagrange multipliers as control inputs, we design two feedback laws that steer the system toward a stationary point of the optimization problem, yielding two novel algorithms with proven global exponential convergence in the strongly convex case. Numerical experiments demonstrate their effectiveness.

ABSTRACT. This abstract summarizes the results in [1,2] where we address state-of-charge (SoC) estimation for lithium-ion batteries modeled by the Single Particle Model (SPM), focusing on the interaction between spatial discretization, physical consistency, and observer design. A nonuniform finite-difference (NU-FDM) discretization of the solid-phase diffusion equation is derived, and the surface-input gain is selected so that the discrete bulk coordinate exactly matches the continuous bulk normalized concentration. The full-cell NU-FDM and finite-volume (FVM) models are shown to be structurally unobservable because the two electrode diffusion matrices contribute two zero eigenvalues to a system with a scalar voltage output. The full-cell mass-conservation relation is then used to eliminate one internal negative-electrode state and reconstruct it algebraically. The positive- and negative-electrode grids are optimized jointly using SoC and terminal-voltage error objectives evaluated with an extended Kalman filter. Numerical results show that reconstruction conditioning strongly affects the uniform-grid baseline. SoC-oriented optimization reduces the estimation objective by up to $86.5\%$, while a normalized combined objective reduces the SoC objective and terminal-voltage RMSE by $84.3\%$ and $71.2\%$, respectively, for the representative case $N=6$, $k_{\mathrm{rem},-}=3$.

ABSTRACT. This paper addresses the distributed solution of integer knapsack problems over time-varying communication networks. We propose a distributed realization of the Bellman recursion in which each agent, associated with a single item, performs local dynamic programming updates while aggregating value estimates through a max-consensus mechanism. We show that the resulting distributed iterations are monotone and bounded, and prove finite-time convergence under joint strong connectivity of the communication graph. Moreover, we establish that the limiting value function coincides exactly with the centralized optimal solution, thereby guaranteeing distributed optimality without approximation. The analysis reveals that the acyclic structure of the Bellman recursion, combined with max-consensus propagation, enables exact decentralized implementation of dynamic programming in discrete combinatorial settings.

ABSTRACT. We consider the problem of moving an open liquid-filled container in minimum time, along a straight line, preventing the liquid from spilling. The container is actuated by a planar motor, that allows controlling three degrees of freedom (horizontal, vertical position and pitching angle). We assume an axisymmetric container and small liquid oscillations, and approximate the liquid dynamic with a damped pendulum. Experimental results confirm the effectiveness of the model and show that the possibility of controlling the tank angles, allowed by the use of a planar motor, allows reducing the transition time, especially when the allowed maximum sloshing is low.

ABSTRACT. Since 2024 the Control Research Group at the University of Siena has had a fruitful collaboration with the Co.Ri.M.A. Division of Marchesini Group. Students completing the internship there, are enabled to tackle practical control problems that arise in cutting-edge applications. This work presents what has been done across three thesis projects on the mitigation of the sloshing phenomenon in automated liquid handling systems. The proposed approach is based on the direct design of the motion laws by means of optimization. In a first thesis project, a rectilinear motion has been considered and a quadratic optimization problem based on a linear equivalent mechanical sloshing model has been proposed to design the motion law. In a second project, a piecewise affine approximation has been used to to capture non-linear dynamics, leading to the formulation of a mixed-integer quadratic program. Finally, in a third project, the approach has been adapted to circular trajectories typical of carousel machines. Additionally, robust and stochastic formulations have been introduced to account for variations in container fill levels. Experimental validations have demonstrated that the optimized motion laws significantly reduce both peak oscillations during movement and residual oscillations in the resting phase compared to industrial polynomial and trapezoidal benchmarks.

ABSTRACT. We propose a model-based activation law for a vehicle’s Electronic Stability Program (ESP). In contrast with to standard industrial practice, which activates the ESP when the error between the measured yaw rate and a reference signal exceeds a dead-zone threshold, we define a safe set in the side-slip angle / yaw rate state space building on nonlinear modelling and bifurcation analysis. Once the safe set is left, stability guarantees are lost and the the ESP is turned on. Testing in a high- fidelity multibody simulator shows improved robustness across manoeuvres, compared to the standard dead-zone law

ABSTRACT. This paper summarizes preliminary results on automated driving. In particular, Time-Varying Artificial Potential Fields (TVAPFs) are defined to achieve joint decision-making (DM) and trajectory planning (TP) by formulating a finite-horizon optimal control problem (FHOCP). In this formulation, maneuver selection, i.e., DM, and feasible trajectory generation, i.e., TP, are solved within the same optimization problem. We first introduce the framework and define TVAPFs for dynamic road actors under bounded prediction uncertainty, enabling collision-aware planning in multi-actor road scenarios. Then, the framework is extended to unsignalized intersections by introducing an actor-selection strategy, predicting short-horizon conflict-zone occupancy, and defining an intersection-specific TVAPF. Simulations in overtaking and intersection scenarios show that behaviors such as delaying, yielding, overtaking, stopping, and proceeding can emerge from the FHOCP without discrete mode switching.

ABSTRACT. This study presents a Digital Twin (DT)-based system to solve the last-mile delivery problem. The city is divided in zones, each zone is associated to a hub where the trucks arrive and the last-mile deliveries are performed by a set of robots. The DT architecture is composed of an application layer that optimize the robot initial routes, a simulation layer connected with the physical layer to manage unexpected events, such as traffic congestion and road closures. A Mixed Integer Linear Programming (MILP) optimization model is designed to determine the optimal routes of the robots and a simulation model is implemented for the rerouting application. The system is validated through two case studies conducted in the city of Bari, Italy, using the MILP model to reduce travel distances and energy consumption of robots and the Simulation of Urban Mobility tool for reproducing the DT physical layer. The results demonstrate the effectiveness and adaptability of the proposed approach.

ABSTRACT. Distributed control of System of Systems (SoS) has recently drawn a lot of attention from the scientific community due to its wide array of real case applications. One of the most important aspects considered when designing a controller is stability checking, in fact, it may be essential to understand whether a control law can be applied. In this work, the concept of consensus horizon is introduced as the possibility of reducing the number of deciding variables that are required to reach consensus, creating a sibling optimization problem with fewer consensus constraints. Then, unconstrained MPC, Consensus ADMM (C-ADMM), and Linear Quadratic Tracking (LQT) techniques are used to solve the sibling optimal distributed control problem and perform a stability analysis of the closed-loop system.

ABSTRACT. In this paper, we propose a generalized framework for the design and analysis of suboptimal and reduced-order nonlinear Model Predictive Control (MPC) architectures. The proposed framework manages real-time operation of MPC schemes by (i) computing the control action suboptimally, i.e., by running a generic optimal control algorithm for a finite number of iterations, and (ii) relying on a reduced-order model that neglects part of the plant dynamics (accounting for, e.g., unmodeled dynamics or a low-level compensator). To rigorously handle the interplay between optimization error and model mismatch, we treat the sampling time as a tunable design parameter. We analyze the resulting closed-loop system, comprising the full-order physical plant interconnected with the iterative optimization algorithm (treated as a dynamical system), by leveraging tools from timescale separation. We prove that operating at a sufficiently fast sampling rate ensures that the closed-loop system maintains recursive feasibility and achieves an exponentially stable equilibrium point. The effectiveness of the proposed framework is validated on an underactuated two-link robotic arm through virtual experiments in the high-fidelity MuJoCo physics engine.

ABSTRACT. We present a novel economic Model Predictive Control (MPC) formulation with reference tracking adaptation. The proposed approach is based on an adaptive performance index which enables a smooth transition from an economic formulation - potentially non-convex and yielding only a suboptimal solution - to a convex tracking one. The adaptive mechanism is regulated by a contraction condition related to the cost function along the closed-loop trajectory. The adaptive policy enables a novel fashion for providing the closed-loop stability of the economic MPC, going beyond standard dissipativity-based and rotated cost arguments that typically rely on strong duality. Indeed, the proposed formulation is particularly appealing when, given an economic criterion, the system is non-dissipative and no stability certificates can be given. Furthermore, the proposed economic MPC retains recursive feasibility and stability throughout the closed-loop evolution even with changes of the economic criterion, which typically might destabilize standard economic MPC controllers.

ABSTRACT. Model Predictive Control (MPC) provides a suitable framework for controlling constrained nonlinear systems since state and input constraints can be explicitly included in the optimal control problem. In the context of manifold stabilization, transverse normal form (TNF) coordinates provide a useful representation in which the motion transverse and tangential to a target manifold can be treated separately. This separation allows the controller to prioritize convergence to the manifold over the motion along it. We propose a compact artificial reference TNF-MPC formulation, where the artificial reference is expressed in TNF coordinates and is optimized as a steady point. The resulting structure combines the geometrical interpretation of TNF-based manifold stabilization with the flexibility of the artificial-reference tracking MPC.

ABSTRACT. This work addresses the trajectory tracking problem for autonomous vehicles by proposing an integrated longitudinal/lateral control strategy based on Linear Parameter-Varying (LPV) Model Predictive Control (MPC). The main objective is to accurately follow a feasible reference trajectory while satisfying state and input constraints and keeping the online optimization time compatible with real-time operation. The proposed LPV MPC uses a control-oriented vehicle model expressed in Frenet coordinates, in which longitudinal speed and a curvature-dependent term are used as scheduling variables. This formulation preserves the coupling between lateral and longitudinal dynamics while reducing the computational burden typically associated with nonlinear MPC. Simulation results on a high-fidelity nonlinear vehicle model show accurate trajectory tracking and a significant reduction of computation time, supporting the suitability of the method for real-time autonomous driving applications.

ABSTRACT. This paper proposes a hybrid optimization–learning framework that integrates the Alternating Direction Method of Multipliers (ADMM) with Reinforcement Learning (RL) to accelerate nonlinear model predictive control (MPC) in real time. The approach addresses the computational bottleneck of iterative optimization by training a reinforcement learning agent to act as a solver-aware meta-optimizer, adaptively selecting ADMM warm-start initialization and penalty parameter based on the current operating condition and solver diagnostics, rather than learning the plant control policy directly. The method is validated on the quadruple-tank plant, a strong coupling multivariable nonlinear system. By leaving the underlying formulation and ADMM update equations unchanged, the proposed RL-enhanced scheme preserves the solution structure of the original optimization problem while significantly reducing the number of iterations required for convergence. Extensive simulations, including Monte Carlo analyses over randomized operating points, demonstrate improved computational efficiency with closed-loop performance comparable to standard ADMM-based MPC. Experimental results on the physical plant further highlight the effectiveness of RL-based meta-optimization as a practical strategy for real-time control of complex nonlinear systems.

ABSTRACT. Tuning Model Predictive Control (MPC) cost weights for multiple, competing objectives is labor-intensive. We present Differentiable Weights-Varying MPC (Diff-WMPC), which learns a look-ahead policy for context-dependent NMPC cost weights using solver sensitivities and end-to-end gradients. The method combines constrained nonlinear MPC with differentiable, gradient-based training, enabling direct optimization of task-level tracking objectives while preserving real-time control structure. On a full-scale racecar simulation model, Diff-WMPC outperforms static-weight baselines and reduces training time from over an hour to under two minutes relative to weights-varying Reinforcement Learning (RL). The learned policy transfers zero-shot to unseen conditions and, with quick online fine-tuning, reaches environment-specific performance. Project website: https://diffmpc.com.

ABSTRACT. Building on our recent hybrid MPC-deep reinforcement learning (RL) smart charging algorithm, in which a neural network (NN) tunes online the terminal cost weights of a model predictive controller (MPC) for plug-in electric vehicles (PEVs), this work-in-progress reports a computation-aware extension. Two separate actors, trained offline with TD3, respectively select the terminal weights and the effective prediction horizon, the latter guided by a reward that explicitly accounts for the MPC solving time. A preliminary campaign identifies the conditions under which each mechanism is effective: the adaptive horizon pays off when the MPC iteration is a mixed-integer problem, where solving time grows steeply with the horizon while control performance saturates; the terminal weights matter when the two control objectives genuinely compete, prioritizing some PEVs over others. In the congested regime, the dual-actor controller attains a 7% lower combined error than the adaptive MPC while reducing the average solving time by 59%.

ABSTRACT. Reactive navigation in unknown and dynamic environments is traditionally addressed by constructing a discrete representation of the free space and planning over the resulting grid, graph, or sample set. This work presents an optimization-based methodological framework in which local range measurements are converted into a continuous and differentiable representation of the search space, so that navigation can be formulated directly as an online optimization problem. To this aim, the framework first yields NAPVIG, a constrained gradient procedure that computes local approximations of the Generalized Voronoi Diagram (GVD) from the discrete measurements; then, it is followed by LOPWEEG, which extends this formulation by extracting and optimizing a local topology of feasible branches for online decision making. The resulting method is map-free, sensor-driven, computationally tunable, and compatible with receding-horizon execution.

ABSTRACT. A hybrid locomotion approach is proposed for a quadrupedal robot to enhance energy efficiency on mixed terrains. A lightweight mechanical modification consisting of four passive castor wheels mounted on the robot’s abdomen enables stable rolling while preserving standard locomotion, with only a 4% increase in robot mass. A neural network was developed to estimate power consumption, trained on real robot data and directly applied in simulation. The neural network achieved a coefficient of determination of R2 = 0.97, ensuring accurate estimation of energy consumption in simulated environments. Experimental and simulated results show that the proposed rolling gait reduces the average Cost of Transport(CoT) from approximately 4.5–6.0 during trotting to 0.8–1.1 during rolling, corresponding to a four–five-fold improvement in energy effi- ciency.

ABSTRACT. Clothoids basically consist of continuous linear curvature trajectories. Since spline of clothoids can be used to model natural trajectories of mobile robots, their tracking is of interest, e.g., in robotic leader-follower settings. To track this class of trajectories, we propose adopting an Internal Model-based controller, which elaborates the tracking error to reproduce the unknown control input necessary to keep the tracking error to zero. In this work, we detail how to specialise a controller of the aforementioned class to the case of a multirotor. In detail, the theoretical proofs are provided for the specific use case, while numerical results demonstrate the performance of the proposed control system.

ABSTRACT. This paper presents a cascade geometric control approach for multirotors with tiltable rotors. The proposed framework enables simultaneous position and attitude tracking while avoiding singularities associated with local attitude representations. By exploiting thrust vectoring, the vehicle can generate lateral forces without significant body tilting. Numerical simulations demonstrate effective tracking performance and robustness to disturbances and actuator imperfections. This extended abstract is based on [1].

ABSTRACT. Recent advances in aerial robotics have enabled multirotor UAVs to perform physical interaction tasks, requiring the generation of desired interaction wrenches while tracking reference trajectories. Conventional multirotors with fixed thrust directions are intrinsically underactuated and cannot independently control arbitrary force and moment vectors. Thrust-vectoring UAVs overcome this limitation by dynamically reorienting propeller thrust directions, enabling full actuation and reshaping the feasible wrench space. However, exploiting these additional actuation capabilities requires efficient control allocation strategies capable of adapting the vehicle configuration while preserving computational efficiency. This work investigates a heuristic allocation strategy for a star-shaped interdependent cant-tilting hexarotor UAV. The proposed approach exploits a geometric polytope-based characterization of the achievable force space to select the cant configuration through a lightweight online procedure, while the rotor speeds are computed through pseudo-inverse allocation. Simulation results in physical interaction scenarios demonstrate that the proposed strategy significantly reduces the computational burden compared to optimization-based allocation while preserving tracking and interaction performance. The results highlight the potential of geometric reasoning for efficient control allocation in reconfigurable multirotor platforms.

ABSTRACT. This paper proposes Super-Twisting sliding mode controllers (STSMCs) to solve the trajectory tracking problem of a cargo Uncrewed Aerial Vehicle (UAV), while actively reducing the oscillations of the cable-suspended payload hanging from the quadrotor. In order to damp the oscillatory behavior of the cargo, we propose sliding surfaces which include the dynamics of the load swing angles. The proof of stability is carried out through Lyapunov analysis, resulting in finite-time convergence of the sliding surfaces and their derivatives, and asymptotic convergence of the tracking errors. Simulations in Matlab and Simulink are presented to show the effectiveness of the designed STSMCs, considering a small-scale cargo drone which performs a circular and a helical trajectory. Under the proposed controllers, the UAV is able to track the references, while the load swing angles do not exceed the magnitude of 2.5 degrees. Our approach allows to reduce load oscillations by more than 50 % when compared to STSMCs without active swing reduction.

ABSTRACT. This paper proposes a formation controller for a fleet composed of two unicycles and a quadcopter. The formation problem is defined as a min-max problem whose optimal control strategy is linear in the local measurements of each agent, and the matrix gains are obtained by solving a Linear Matrix Inequality. Artificial repulsive forces are added to avoid inter-agent collisions and obstacles. The proposed control scheme was validated both in simulation and through several experiments, involving both static and dynamic obstacles. The results show that the agents can achieve the formation without crashes.

ABSTRACT. Agricultural robotics has recently attracted increasing interest in reactive, map-free navigation strategies capable of operating reliably in semi-structured environments. This paper presents and compares two LiDAR-based autonomous navigation approaches designed for ground rovers operating within orchard rows. Both methods rely exclusively on real-time 2D LiDAR data, avoiding the need for prior maps or external localization systems. The proposed strategies are formulated in a general framework and implemented on two rover platforms with different kinematic configurations, namely a differential drive and a double steering axle vehicle. A comparative evaluation is conducted through real-world experiments, highlighting the strengths and limitations of each strategy in terms of accuracy, robustness, and adaptability to different vehicle models.

ABSTRACT. We present a global optimization approach for nonlinear regression, based on an interpolation-based convex relaxation. We approximate the image of a nonlinear map through a convex polytope obtained by multilinear interpolation over a grid, providing a convex underestimator of the original nonconvex objective function. The resulting relaxation yields a tight and computationally tractable lower bound, whose accuracy depends explicitly on the second derivatives of the nonlinear mapping, and can be efficiently embedded within a Branch-and-Bound (BnB) algorithm to guarantee global optimality. Theoretical results establish error bounds for the interpolation based approximation, and characterize the convergence properties of the derived relaxation. We apply the method to the identification of the parameters of the pharmacokinetic/pharmacodynamic (PK/PD) model of propofol, a drug used in general anesthesia.

ABSTRACT. We present a unified framework for identifying nonlinear dynamical systems in which a physics-based parametric model is embedded with a data-driven kernel term that accounts for unmodeled dynamics. A generalized representer theorem yields the joint estimate of interpretable physical parameters and of the nonparametric correction as the solution of a single regularized problem, with a closed form in the frequent affine-in-parameters case and a state-space extension based on nonlinear smoothing for partially measured states. Building on this framework, worst-case multi-step error bounds isolate the role of the predictor's Lipschitz constant, which nonexpansive kernels render available in closed form. This turns the regularization weight into a design variable, enabling a Lipschitz-based identification that shapes error propagation during training. The approach is validated on academic and experimental benchmarks.

ABSTRACT. This study compares classical Sparse Identification of Nonlinear Dynamics (SINDy), using sequential thresholded least squares in a single global fit, with Ensemble SINDy, which applies bootstrap resampling, degree-tiered greedy term selection, a redundancy guard, inclusion-probability thresholding, and debiased refitting. Both methods share identical state data, finite-difference derivatives, and candidate libraries, differing only in coefficient selection. Using an LQR-stabilized reaction wheel pendulum, candidate library size was varied as H = 4, 35, 50, and 70, the largest including trigonometric terms. Both methods show comparable coefficient error for small libraries, but classical SINDy degrades sharply as library size grows, since a single trajectory cannot sufficiently excite the enlarged, collinear basis. Ensemble SINDy maintains near exact coefficient recovery across all sizes, with its H = 70 trajectory reconstruction closely matching measured data, while the classical model fails. Ensemble-based selection thus offers markedly greater accuracy and robustness for large, over complete libraries, with equivalent performance on small, well-conditioned ones. The study extends next to the nonlinear pendulum and Chua's circuit.

ABSTRACT. Generalized Lotka–Volterra (gLV) models are systems of nonlinear ordinary differential equations widely used to describe interacting microbial populations. However, parameter estimation can be challenging when experimental uncertainty affects the inferred coexistence properties, while the estimated parameters must simultaneously satisfy feasibility and stability conditions. We present a stability-aware identification framework that combines stationary and time-series data with analytical feasibility and stability conditions. Stationary measurements are used to estimate normalized growth and interaction parameters, while time-series data are used to estimate the self-limitation coefficients and complete the model identification. When stationary measurements are inconsistent with experimentally observed coexistence, log-normal noise is introduced to generate perturbed datasets. The framework is applied to an in vitro reconstructed seven species microbial community. The results show that small variations in stationary measurements may move the inferred system across a stability boundary, leading to qualitatively different coexistence predictions. We also consider a modified gLV model with Michaelis–Menten-like saturating interactions, showing how saturation changes the stability condition and enlarges the parameter region compatible with stable coexistence. Overall, the proposed approach combines parameter estimation and qualitative analysis to recover models consistent with both experimental observations and the expected dynamical behavior of the system.

ABSTRACT. We present an overview of the approach proposed in [1] to compute stabilizing controllers for continuous-time linear time-invariant systems directly from an input–output trajectory affected by process and measurement noise. The proposed output-feedback design combines (i) an observer of a non-minimal realization of the plant and (ii) a feedback law obtained from a linear matrix inequality (LMI) that depends solely on the available data. Under a suitable interval excitation condition and knowledge of a noise energy bound, the feasibility of the LMI is shown to be necessary and sufficient for stabilizing all non-minimal realizations consistent with the data.

ABSTRACT. The dynamic control allocation problem for LTI systems is addressed in an uncertain setting, using a completely data-driven strategy in the presence of non-constant steady-state behavior, while leaving untouched the regulated output response induced by an a priori given controller. Compared with the current state of the art, the proposed solution exhibits several appealing features: complete invisibility of the allocator’s action (after a training interval if the plant is unknown), exact optimization of the periodic steady-state evolution, arbitrary speed of the allocation action.

ABSTRACT. We establish finite sample certificates on the quality of solutions produced by data based forward-backward (FB) operator splitting schemes. As frequently happens in stochastic regimes, we consider the problem of finding a zero of the sum of two operators, where one is either unavailable in closed form or computationally expensive to evaluate, and shall therefore be approximated using a finite number of noisy oracle samples. Under the lens of algorithmic stability, we then derive probabilistic bounds on the distance between a true zero and the FB output without making specific assumptions about the underlying data distribution. We show that under weaker conditions ensuring the convergence of FB schemes, stability bounds grow proportionally to the number of iterations. Conversely, stronger assumptions yield stability guarantees that are independent of the iteration count. We then specialize our results to a popular FB stochastic Nash equilibrium seeking algorithm and validate our theoretical bounds on a control problem for smart grids, where the energy price uncertainty is approximated by means of historical data

ABSTRACT. Autonomous robotic systems are increasingly required to translate high-level mission objectives into executable task sequences while accounting for resource constraints, heterogeneous capabilities, and limited communication. In this context, the task manager plays a central role, deciding what should be done, when, and by which agent, while preserving real-time feasibility and operational robustness. This work presents a unified learning-based task management framework composed of two reinforcement learning modules. A Double Deep Q-Learning module addresses resource-aware sequential decision-making on graph-structured environments. A Multi-Agent Proximal Policy Optimisation module extends this reasoning to decentralised task allocation, where heterogeneous agents coordinate through lightweight, human-like messages encoding task intention, execution cost, and expected contribution. The resulting task manager operates across two complementary levels: individual resource-aware execution and team-level decentralised allocation. Finally, explainability-driven policy analysis is introduced as a diagnostic layer to inspect learned behaviours, reveal hidden coordination failures, and guide reward or architecture redesign. The proposed perspective integrates the authors’ recent contributions into a technical trajectory toward interpretable, scalable, and adaptive task management for autonomous robotic systems. A climate-resilient environmental monitoring scenario is presented as a representative case study to ground the framework in a real-world application.

ABSTRACT. This paper summarizes the experimental results obtain with a Model Predictive Control (MPC) strategy for automatic regulation of Depth of Hypnosis (DoH) during Total Intravenous Anesthesia (TIVA). Unlike previously developed solutions based on Proportional-Integral-Derivative (PID) controllers, MPC allows patient demographic information to be embedded in the control law and clinical constraints to be handled explicitly. We describe the main features of a Generalized Predictive Control (GPC) scheme based on an external predictor, and we present in-vivo results obtained on four patients undergoing elective plastic surgery. The outcomes confirm the clinical feasibility of the approach while also revealing residual performance limitations that are shown to be closely related to the uncertainty on the pharmacological model. This observation motivates the ongoing extension of the control architecture toward an adaptive scheme in which the model is periodically updated through robust identification of patient-specific parameters.

ABSTRACT. We present a system-theoretic perspective based on passivity tools for Tracking-ADMM, a recently introduced distributed optimization algorithm for constraint-coupled optimization problems. We formulate the algorithm as a dynamical linear component in closed-loop with a static nonlinearity resulting from an optimization step, which allows us to assert algorithmic convergence in terms of absolute stability of the interconnection.

ABSTRACT. Classical fixed-point theorems guaranteeing the existence of equilibria in noncooperative games, such as those of Brouwer and Kakutani, rely on convexity assumptions. Nevertheless, many games are inherently nonconvex, rendering such results inapplicable. In this work, we establish the existence of equilibria in games by employing tools from algebraic topology. We introduce a local equilibrium notion, the first-order Nash equilibrium, that captures stationary strategy profiles. Using the Lefschetz fixed-point theorem, we prove that at least one local equilibrium exists under mild assumptions, even when cost functions are nonconvex. We further refine this result through the Nielsen extension of the Lefschetz theorem, which characterizes the persistence of equilibria under continuous deformations of the game.

ABSTRACT. This work studies feedback Nash equilibria (FNE) in scalar discounted linear quadratic (LQ) games with N players. By explicitly incorporating the discount factor, we show that finite-cost equilibria may fail to stabilize the original system, motivating a distinction between FNE and stable FNE together with a sufficient stability condition. Based on a parametric characterization of the policies, we propose numerical methods for computing all equilibria. Particular attention is devoted to the symmetric game, where a closed-form expression of the symmetric FNE and conditions for the existence of up to $M\leq2^N-2$ equilibria are derived. Numerical experiments illustrate how equilibrium multiplicity depends on the game configuration and highlight the emergence of finite-cost non-stabilizing equilibria.

ABSTRACT. Social systems are often characterized by the presence of opinion formation processes and decision-making processes that often co-evolve, mutually influencing one another. In fact, an individual's decision on which action to take is often influenced by their opinion, which in turn is affected by the behaviors observed from others. The co-evolutionary model of actions and opinions captures these mechanisms by means of a game-theoretic model; individuals interact on a network and simultaneously revise their action and opinion following a best-response dynamics. Building on this model, we formulate and study the problem of optimally targeting the network to guide the system from a consensus on an action to a different one. In particular, we consider the scenario in which a set of stubborn agents (committed minority) can be introduced, using social influence to drive the rest of the population to the desired consensus. For this problem, we derive a polynomial-time algorithm to determine whether targeting a specific set is sufficient to reach the desired objective and, building on this result, we derive an effective heuristic to solve the optimal placement problem, identifying the minimal target set.

ABSTRACT. Complete sub-optimal MAPF solvers are widely used thanks to their ability, even in crowded scenarios, to find a feasible solution that brings each agent to its target, preventing deadlock situations. However, generally, these algorithms provide much longer solutions than the shortest one. Our main contribution is a new local search procedure for improving a known feasible solution. We start from a feasible sub-optimal solution, and perform a local search in a neighborhood of this solution. If we are able to find a shorter solution, we repeat this procedure until the solution cannot be shortened anymore. At the end, we obtain a solution that is still sub-optimal, but generally of much better quality than the initial one. We propose two different local search policies. In the first, we explore all paths in which the agents positions remain in a neighborhood of the corresponding positions of the reference solution. In the second, we set an upper limit to the number of agents that can change their path with respect to the reference solution. These two different policies can also be alternated. We explore the neighborhoods by dynamic programming. The fact that our search is local is fundamental in terms of time complexity.

ABSTRACT. In industrial environments shared by autonomous mobile robots and human operators, fleet coordination requires multi-agent plans that are free of conflicts and consistent with the footprint and kinematics of the vehicles. When unmapped obstacles appear during execution, systematic global replanning rebuilds the entire plan even for localized perturbations. This work integrates a three-stage incremental replanning pipeline, based on the incremental search algorithm DLPA*, into a volume-aware Conflict-Based Search framework, and compares it with global replanning under identical conditions. Repairs remain confined to the affected agents, the quality of the repaired plans stays close to that of global replanning, and the search effort is reduced in most replanning episodes.

ABSTRACT. The design of state observers for cyber-physical systems under sparse sensor biases, arising from either malicious attacks or faults, has drawn substantial attention in recent years. While batch approaches are well-established, the development and analysis of online observers remain open problems. In this work, we analyze observability and study online sparse observers derived from online sparse convex optimization and block Bregman methods. In particular, we discuss their convergence properties and we compare their performance through numerical experiments.

ABSTRACT. In this paper, we tackle the problem of ensuring string stability in connected autonomous vehicle (CAV) platoons under sampling, quantization, and external disturbances. Building on prior work that examined the impact of sampled-data control and quantized information exchange on platoon stability, we extend the analysis to scenarios where microscopic (i.e., CAVs' positions, speeds, and accelerations) and macroscopic (i.e., traffic flow information) measurements are asynchronous. The proposed control uniquely accommodates both Disturbance String Stability and Practical String Stability. The proposed method is validated through simulations, highlighting its efficacy in realistic vehicular environments.

ABSTRACT. Deploying drone swarms to track a dynamic target in cluttered environments presents severe computational and safety challenges. We propose MRoPE, a hierarchical strategy that decouples the cooperative monitoring mission from strict local safety requirements. To overcome the computational bottlenecks typical of dense spaces, our approach dynamically aggregates complex obstacle geometries into a single safe bounding ellipse for each drone. Methodologically, this architecture is realized by combining distributed aggregative optimization for high-level swarm coordination, a decentralized consensus scheme for the safe area computation, and local Predictive Safety Filters (PSF) for real-time collision avoidance. Virtual and real-world experiments validate the framework, demonstrating superior real-time efficiency and scalability compared to centralized approaches.