ABSTRACT. Many relevant problems in multi-agent systems (MAS) such as formation control and coordination can be addressed with different solutions, including either centralized or distributed approaches, protocols based on either absolute or relative information, and many others. However, most of the related works address these topics only at a theoretical level, without investigating the practical applicability of the proposed control strategies. The aim of this paper is to showcase our recent works on the implementation of multi-agent systems control with an experimental setup based on a set of commercially available micro-robots and a vision-based position tracking system. With such a setup we were able to test and characterize face-to-face interaction dynamics and synchronization of moving chaotic oscillators in a real yet controllable multi-agent robotic implementation.

ABSTRACT. This extended abstract addresses the proposal of a new spatial sampling algorithm for encoding geometric information through arc-length parametrization in demonstrated trajectories.

ABSTRACT. The advent of soft robotics has greatly changed researchers’ perspectives on robotic manipulation. The philosophy of embedding the robot intelligence at the design level has led to new prototypes of robotic hands that show capabilities inconceivable with their rigid counterparts. This contribution outlines on-going research efforts on how to unlock new grasping skills in soft-rigid grippers exploiting the interaction with the environment.

ABSTRACT. Visual planning methods are promising to handle complex settings where extracting the system state is challenging. None of the existing works tackles the case of multiple heterogeneous agents, characterized by different capabilities and/or embodiment. We propose a centralized method to realize visual action planning in multi-agent systems (MASs) by exploiting a roadmap built in a low-dimensional latent space and used for planning. To enable MASs, we infer possible parallel actions from a dataset composed of tuples associated with individual actions. Next, we evaluate feasibility and cost of them based on the capabilities of the MAS and endow the roadmap with this information. The approach is validated in a simulated burger cooking task and a real-world box packing task.

ABSTRACT. This work proposes a distributed estimation-control scheme for the coordination of two heterogeneous robots: a steering car and a quadcopter. The car, designated as the leader, must follow a desired trajectory while maintaining a safe distance from static obstacles. The quadcopter follows the car from above to maintain formation. To address this problem, a Nonlinear Model Predictive Controller (NMPC) is combined with a Nonlinear Moving Horizon Estimator (NMHE). While various studies have demonstrated the robustness and accuracy of NMPC-NMHE schemes in different applications, this is the first application to such a heterogeneous system. Unlike classical control techniques, the presented scheme can handle noisy, partial, and missing measurements of the agents' states. The simulation results are promising and suggest further exploration in this direction. Thanks to reliable estimates from the NMHE and accurate predictions from the NMPC, the ground car successfully avoids detected obstacles while keeping the tracking errors of both robots within a few centimeters.

ABSTRACT. We present a neuro-inspired control framework for autonomous robots that integrates an artificial emotion of fear, drawing inspiration from LeDoux’s dual-pathway hypothesis. To replicate the ”Low Road” pathway, our system comprises proxies for the thalamus, implemented as a nonlinear filter; the amygdala, modeled as a Soft Actor-Critic (SAC) reinforcement learning agent; the brainstem, orchestrated through a Nonlinear Model Predictive Controller (NMPC). The NMPC’s parameters are adjusted by the amygdala, enabling it to generate control inputs to actuate the robot. Our preliminary results demonstrate that the robot exhibits a better adaptive behavior than a standard NMPC in both static and dynamic environments with obstacles characterized by different hazard levels.

ABSTRACT. Electric Vehicles (EVs) are emerging as a promising, cleaner and more sustainable solution future for transportation, aiming to combat climate change and reduce our dependence on fossil fuels. EVs powertrain plays the main role in determining the overall performance and efficiency of an electric vehicle. An EV powertrain is composed of several interconnected subsystems, including the electric motor, power electronics (usually a DC/AC converter), and the energy storage system (usually lithium-ion batteries). Efficiently managing and optimizing these subsystems is vital to achieve superior vehicle performance, extended driving range, and enhanced energy efficiency. In recent years, substantial advancements have been made in powertrain control strategies and optimization techniques, owing to breakthroughs in technology, computing power, and an increasing understanding of EV dynamics. The implementation of sophisticated control algorithms, intelligent energy management systems, and real-time optimization has revolutionized the way electric vehicles perform on the road. In this work, the purpose is to leverage the nonlinear Model Predictive Control (MPC) framework in order to design an efficient Adaptive Cruise Control (ACC) which accounts for the EV powertrain dynamics, in order to optimally managing the energy consumption of the battery to increase its range, and, thus, the energetic efficiency of the powertrain as a whole. A classic car following scenarios is studied, simulating a realistic ACC case, with the ego vehicle being controlled by the MPC, and the reference being generated using a Constant Time Gap (CTG) policy. Finally, the MPC is used to generate the required torque for a complete high-fidelity model of the case study vehicle, a Fiat 500e.

ABSTRACT. Model Predictive Control~(MPC) has been commonly used in autonomous driving. While it is well-known how to design an MPC controller such that closed-loop stability w.r.t.~a reference trajectory (or path) is obtained and that constraint satisfaction holds at all times, the results in the literature rely on assumptions that are violated in practice. We address the problem of ensuring the safe operation of an autonomous vehicle in uncertain environments by relying on the Model Predictive Flexible trajectory Tracking Control~(MPFTC) framework, and we show in practice how a safe vehicle controller can be designed for urban autonomous driving settings.

ABSTRACT. We consider a road network represented by a directed graph. We assume to collect many measurements of traffic flows on all the network arcs, or on a subset of them. We assume that the users are divided into different groups. Each group follows a different path. The flows of all user groups are modeled as a set of independent Poisson processes. Our focus is estimating the paths followed by each user group, and the means of the associated Poisson processes. We present a possible solution based on a Dynamic Programming algorithm. The method relies on the knowledge of high order cumulants. We discuss the theoretical properties of the introduced method. Finally, we present some numerical tests on well–known bench- mark networks, using synthetic data.

ABSTRACT. The paper proposes a decentralized optimization approach based on opinion dynamics, for decision making in multi robot systems, during trajectory planning in dynamic environments to avoid deadlocks and indecision's.

ABSTRACT. This work investigates an alternative approach to current control systems for the Automated Driving (AD) of shuttle vehicles on dedicated roads. The proposed solution decouples the problem into two levels: a Deep Deterministic Policy Gradient (DDPG) Reinforcement Learning (RL) agent and a dedicated vehicle logic generating Virtual Lane (VL) data to eliminate redundancy and allow for smooth lane changes on curved roads. The training uses an environment defined through a model-based simulation, exploiting MatLab and Simulink tools, and has been conducted following a Curriculum Learning strategy. The performance of the introduced approach have been evaluated by testing the agent capabilities and exploring its behavior in the presence of external disturbances in the controlled states.

ABSTRACT. This extended abstract addresses the modeling and simulation of a Steering Vehicle Differential for automotive applications.

ABSTRACT. In the pathway towards fully autonomous spacecraft, space systems shall be designed to operate in an unpredictable and partially unknown environment. In this framework, the problem of sensor fusion is typically handled exploiting Kalman-like filtering techniques. A valid alternative when dealing with \textit{non-Gaussian} probabilities are the so-called Particle Filters (see e.g., [1]), which use a set of samples to represent the posterior distribution of a stochastic process given the noisy and/or partial observations. The Unscented Kalman FIlter (UKF) belongs to the KFs family. However, it is based on a sigma-point method that shares some similarities with the PFs. Indeed, analogously to Kalman filters, the state distribution is represented by Gaussian random variables. However, UKF uses a minimal set of carefully chosen sample points (called sigma points), similar to PF samples, that properly capture the true mean and covariance of the states and are propagated through the nonlinearity.

To deal with non-Gaussian uncertainties and outliers, it is possible to extend the UKF scheme relying on the Huber method and obtaining Huber-based UKF (HUKF). Specifically, the Huber approach is a generalization of the maximum likelihood approach [2] and is based on the combination of minimum $\ell_1$– and $\ell_2$–norm estimators to improve robustness against deviations from the commonly assumed Gaussian probability density functions, thus making the HUKF able to solve non-Gaussian distribution problems efficiently. In detail, the standard cost function, defined as the sum of squared residuals $e_i$, is replaced by an indicator functional $J=\sum_{i=1}^{m}\rho(e_i)$, where $\rho(e_i)$ is a blend of the $\ell_1$ and $\ell_2$ norm functions, such that if $|e_i|\leq \gamma$ than $\rho(e_i)= \frac{1}{2}e_i^2$, otherwise $\rho(e_i)= \gamma|e_i|-\frac{1}{2}\gamma^2$. Here, the threshold parameter $\gamma$ has a great impact on filtering precision and robustness. Hence, for $\gamma\rightarrow 0$, the robustness effect increases as more observations are treated as probable outliers, whereas for $\gamma\rightarrow \infty$ the Huber function behaves like a least square minimization algorithm and the HUKF as a standard UKF. While the HUKF is able to reduce the contribution of uncertain measurements in the computation of the estimator, on the other side this results in an increased variance of the estimator. Indeed, because the corresponding weighting function $\psi(e_i)$ is such that $\psi(e_i)=1$ for $|e_i|\leq \gamma$ and $\psi(e_i)=\gamma/|e_i|$ otherwise, the measurement noise variance $\sigma_i^2$ is modified accordingly. Hence, as the residual $e_i$ grows larger than $\gamma$ so it does its variance as it takes into account the probability that the measurement is an outlier. Vice versa, for $e_i\leq \gamma$, the observation is expected to follow the assumed (Gaussian) distribution, and the variance remains unchanged.

In this abstract, we propose a Huber-based Unscented Kalman Smoother (HUKS), combining the benefits of a HUKF-based filtering techniques for nonlinear systems able to handle non-Gaussianity similarly to PFs with an optimal smoother for balancing the effects on the measurements variance. Specifically, we integrate an optimal smoother into the filtering scheme to incorporate also a memory characteristic, thus improving the state estimates and error covariances by harnessing information contained in the entire available data set. In this way, we obtain a forward Huber-based UKF that processes the measurements, followed by a backward smoothing step over past data and given future observations, which reduces the measurement variance. Indeed, in autonomous rendezvous and proximity operations scenarios, adaptive filtering can help mitigate numerical sensitivity and inconsistent filter performances [3]. In this context, applying smoothing enhances the filter's adaptability and improves state estimation accuracy [4]. Additionally, integrating the Huber approach addresses the limitations of assuming Gaussianity, which is critical given that typical sensor systems. Developing a HUKS can be a significant step towards understanding the benefits of integrating these solutions.

The efficacy of the proposed approach is validated in simulation for an Earth-observation scenario involving a small satellite equipped with commercial-off-the-shelf sensors. In particular, we assume that the GPS sensor has a probability p of providing accurate position measurements y, which adheres to the expected Gaussian distribution for the measurement noise, and a probability of 1-p to fail, resulting in inaccurate measurements (outliers) that deviate from the expected distribution for measurement noise. Figure 1 highlights the impact of the parameter $\gamma$ on the HUKF errors as the probability p varies. Specifically, we observe that lower values of $\gamma$ result in reduced errors due to the filter's enhanced ability to reject outliers. However, this comes at the expense of an increment of the associated standard deviation. Conversely, higher values of $\gamma$ lead to a higher likelihood of accepting outliers as valid measurements, resulting in increased errors, especially in cases where the probability of sensor failures is high. These trends are also confirmed by the RMSE reported in Table 1.

Figure 2 compares the estimated positions using the Huber and the smoother for p=1, i.e., no sensor failure, and p=0.5. The Huber-based filters, i.e. HUKS and HUKF (first row), provide superior estimation accuracy compared to classical filters, such as UKS and UKF (second row). This improvement is particularly evident when p=0.5 (right column), as the higher probability of sensor failure leads to a greater number of outliers, which significantly degrade the predictions of UKF and UKS. As it can be noticed, the adoption of the Huber approach results in increased conservativeness and, consequently, higher standard deviations. However, we observe how the introduction of a smoother can mitigate this effect, thus compensating for the higher variances introduced by the Huber-based approaches.

Last, Table 1 presents the RMSE and the average execution time for each tested filter. Specifically, we can observe that the small increase in computational time associated with the combinations of Huber-based and smoother techniques represents a trade-off for the improved robustness against outliers.

References [1] Cesare Donati, Martina Mammarella, and Fabrizio Dabbene. Single-state weighted particle filter with application to earth observation missions. IFAC-PapersOnLine, 56(2):6024–6029, 2023. [2] Peter J Huber. Robust estimation of a location parameter. In Breakthroughs in statistics: Methodology and distri- bution, pages 492–518. Springer, 1992. [3] Joshua Sullivan and Simone D’Amico. Adaptive filtering for maneuver-free angles-only navigation in eccentric orbits. In 27th AAS/AIAA Space Flight Mechanics Conference, 2017. [4] Cory T Fraser and Steve Ulrich. Adaptive extended kalman filtering strategies for spacecraft formation relative navigation. Acta Astronautica, 178:700–721, 2021.

ABSTRACT. This paper introduces a novel Reinforcement Learning-based controller for autonomous glycemic regulation in the treatment of type 1 diabetes, building on the Deep Deterministic Policy Gradient algorithm. The proposed control method is validated through in-vitro simulations on the Bergman glucoregulatory model, proving that it successfully preserves healthy values of blood glucose concentration, while overcoming both standard clinical practice and classical model-based control techniques in terms of both control effort and computational efficiency for real-time applications.

ABSTRACT. Bacterial quorum sensing (QS) is a cell-to-cell communication mechanism through which bacteria share information about cell density, and tune gene expression accordingly. Pathogens exploit QS to orchestrate virulence and regulate the expression of genes related to antimicrobial resistance (AMR). We compare two synthetic QS circuit architectures: in the first, a single positive feedback loop autoinduces the synthesis of the signal molecule; the second includes an additional positive feedback loop enhancing the synthesis of the signal molecule receptor. Our comprehensive analysis of the two systems and their equilibria highlights the differences in the bistable and hysteretic behaviors of the alternative QS structures. An extensive sensitivity analysis is performed to highlight how parameter variations may lead to phenotype alterations in the system behavior. Building on our sensitivity analysis, we mathematically model four distinct QS inhibition strategies - signal molecule degradation, pharmaceutical inhibition, CRISPRi, and RNAi - which lead to the design of Quorum-Quenching (QQ) therapeutic approaches. Despite the underlying complex mechanisms, we demonstrate that the effect of the proposed QQ strategies on QS can be captured by varying specific parameters within the QS models. Finally, numerical analysis predicts their effect on the steady-state behavior of the two different QS models, revealing critical parameter thresholds that guarantee an effective QS suppression.

ABSTRACT. Bacterial Cellulose (BC) is a promising biodegradable biopolymer synthesized by bacteria in a low energy consumption process based solely on renewable materials (a sucrose-based culture is required). BC-based composites exhibit excellent mechano-electrical transduction properties. However, there has been no systematic optimization of BC-based composites for the realization of transduction or low-cost electronics on BC. In this study, we propose different modeling paradigms categorized according to their basic principles: first principle white-box, grey-box, and black-box models. These paradigms have different levels of detail, interpretability and explainability, which are crucial for understanding model dynamics. The overall mltiphysics white-box model integrates electrical, mechanical and chemical effects into a unique solution, focusing on the coupling factors between the physics domains. In the case of grey-box models, reduced models are considered, while symbolic regression can be employed as enhanced black box models. The dynamic behavior of the BC model can be validated by data-driven characterization using experimental data.

ABSTRACT. In this work we address the problem of tumour growth control by properly exploiting a low-dimensional model that grounds on the Chemical Reaction Network (CRN) formalism. Originally conceived to work both in deterministic and stochastic frameworks, it is shown that, except for the case of very low number of tumour cells, the deterministic approach is appropriate to characterize the system behaviour, especially for control planning purposes. Two alternative control approaches are here investigated. One trivially assumes a constant infusion of external drug administration, the other is designed according to a state-feedback control scheme, with complete or partial knowledge of the state. Pros and cons of both control laws are investigated, showing that the tumour size at the beginning of the therapy plays a role of paramount importance for fixed infusion therapies, whilst only state-feedback laws can eradicate arbitrarily large tumours.

ABSTRACT. The channel gains between user equipaments (UEs) and base stations (BSs) have a fundamental importance when optimizing mobile networks efficiency. In general, these parameters can be evaluated by means of site specific ray tracing solvers that simulate the radio waves propagation. These strategies (or any other relying on physics-based models) have the limitation of depending on how accurate the environment is described and reconstructed into the propagation solvers/models. However, even if the buildings geometry and electromagnetic properties of the construction materials are known, parameters such as motion of objects (people, cars, etc) make these quantities stochastic in practice. In real life situations the electromagnetic waves propagated by the antennas will suffer not only from shadowing, but also combined effects of reflection and diffraction. Consequently, the channel gain between a UE/BS pair is inherently a random quantity. In this paper, we extend the work presented in [1] to the case where the channel gains are random variables. We propose a joint optimization algorithm that minimizes the BSs transmission powers in OFDMA heterogeneous networks, while respecting chance-constrained individual users quality of service (QoS) requirements. Unlike many works in the literature, our optimization problem does not rely on iterative/sequential procedures and does not require a known feasible initial solution, which in general leads to local optimal.

ABSTRACT. Multi-robot systems have gained increasing interest across various fields such as medicine, environmental monitoring, and more. Despite the evident advantages, the coordination of the swarm arises significant challenges for human operators, particularly concerning the cognitive burden needed for efficiently controlling the robots. In this study, we present a novel approach for enabling a human operator to effectively control the motion of multiple robots. Leveraging a shared control data-driven approach, we enable a single user to control the 9 degrees of freedom related to the pose and shape of a swarm. Our methodology was evaluated through an experimental campaign conducted in simulated 3D environments featuring a narrow cylindrical path, which could represent, e.g., blood vessels, and industrial pipes. Subjective measures of cognitive load were assessed using a post-experiment questionnaire, comparing different levels of autonomy of the system. Results show substantial reductions in operator cognitive load when compared to conventional teleoperation techniques, accompanied by enhancements in task performance, including reduced completion times and fewer instances of contact with obstacles. This research underscores the efficacy of our approach in enhancing human-robot interaction and improving operational efficiency in multi-robot systems.

ABSTRACT. Brain-Computer Interface (BCI) systems have gained significant attention as a means of establishing direct communication between the human brain and external devices. Such direct communication provides novel ways of interacting with both healthy individuals, and more importantly, patients with disabilities resulting from spinal cord injuries, Amyotrophic Lateral Sclerosis (ALS), stroke, or other neurological conditions. Electroencephalographic (EEG) recordings of brain waves allow for this kind of connection to be established. In the robotics field, BCI applications have shown promise for intuitive and efficient control. In this context, this paper presents a hybrid BCI approach for controlling a robotic system, combining Motor Imagery (MI) and Error-Related Negativity (ERN) paradigms. The proposed framework intends to use MI to empower users in actively controlling the robot and ERN for human visual feedback to empower users to provide feedback on the robot's state changes. In MI-based BCI systems, users imagine performing a specific movement without actual execution. In particular, when subjects imagine left or right hand, arm, or foot movements, a lateralization can be seen in the data, particularly when the frequency EEG band [7-30 Hz] is considered. By contrast, ERN is a type of error-related potential (ERP) present in the EEG signals when the user is aware of an erroneous behavior. They are characterized by an early negative voltage deflection over fronto-central regions, referred to as error-related negativity (N_E), followed by a positive deflection over parietal regions, referred to as error-positivity (PE). Such peaks occur between 50 and 100 [ms] after the commission of a speeded motor response that the subject immediately realizes to be an error. By combining these two distinct paradigms, i.e. MI and ERN, for active control and feedback retrieval, the BCI framework presented here becomes hybrid in nature. Separate and offline demonstrations, intended as a calibration stage, of the proposed BCI approach's feasibility have been made through simulations and experimental evaluations involving human participants and an Unmanned Aerial Vehicle (UAV). Such a calibration phase results essential as the initial step towards the implementation of active and closed-loop control for BCI systems.

ABSTRACT. Traditional robot programming requires skilled operators, contrasting with small and medium-sized enterprises lacking robotics expertise. Programming by Demonstration offers a viable solution, allowing users to program a robot in a no-coding fashion. However, while the user demonstrates what a skill is about, the robot depicts it only as a sequence of movements. Therefore, the knowledge gathered from the demonstration can hardly be transferred to a different task. This abstract presents the outcomes of enriching Programming by Demonstration to encode the meaning of the skill being demonstrated. We enable the robot to understand the skill semantics in terms of preconditions and effects from a single demonstration. A symbolic planner can then use the taught skills as modular blocks to autonomously compute the sequence of actions required to solve a task. Finally, given the relevance of multi-robot systems on factory grounds, the abstract outlines the guidelines for extending the proposed teaching methodology to multi-agent systems.

ABSTRACT. Supernumerary robotic limbs (SRLs) can restore lost motor functions and enhance human sensorimotor capabilities. An essential challenge is formulating intuitive augmentation policies for controlling SRLs without impeding natural limb functionality. This work introduces an innovative strategy utilizing the redundancy of the human kinematic chain for commanding SRLs having one degree of freedom. This concept is summarized in the definition of the Intrinsic Kinematic Null Space (IKNS). The newly developed procedure encompasses a real-time analysis of body motion and a subsequent computation of the control signal for SRLs based on the IKNS for single-arm tasks. What sets our approach apart is its explicit emphasis on incorporating user-specific biomechanical and physiological characteristics and constraints. This ensures an efficient and intuitive approach to commanding SRLs, tailored to the individual user’s needs.

ABSTRACT. The Metaverse is an immersive shared space that remote users can access through virtual and augmented reality interfaces, enabling their avatars to interact with each other and the surrounding. While digital objects can be manipulated, physical objects cannot be touched, grasped, or moved within the Metaverse due to the absence of appropriate interfaces. This work proposes a solution to overcome this limitation by introducing the “Physical Metaverse”, a shared environment populated by “Avatarms”: avatars equipped with robotic arms capable of performing physical manipulative tasks. These robotic arms remain hidden from the user’s view through diminished reality techniques, enabling users to tangibly perceive the manipulated objects and gain a heightened sense of situational awareness.

ABSTRACT. This study develops a robust optimization model for task planning of human-robot collaborative operations. Our primary objective is to minimize the total completion time of the tasks while taking into account feasibility and operational constraints. The proposed model provides robustness against the uncertainty of human behavior, while allowing the problem to remain deterministic and linear in nature. The level of conservativeness can be adjusted to provide a trade-off between optimality and robustness, such that over-conservatism is avoided in different operational scenarios. The performance and advantages of the model are discussed to highlight its capability to achieve an effective task plan, manage uncertainties, and fulfill constraints in a human-robot collaborative environment.

ABSTRACT. Existing literature on model-based filter design for stochastic LTI systems assumes complete correspondence between the system and its model. When the system is not completely known, the standard indirect model-based (two-steps) filtering solution consists of: (i) identify a model of the system from measured input/output data; (ii) design a Kalman filter based on the estimated model. The performance of this indirect approach are limited by the model and noise covariance matrices accuracy. To overcome such limitations, this paper investigates a direct (one-step) solution to the filtering problem for SISO LTI systems in the Prediction Error Method (PEM) identification framework. Simulation results indicate the effectiveness of the direct filtering approach, especially when the noise covariance matrices are misspecified.

ABSTRACT. This paper addresses the stabilization of a particular class of continuous-time systems affected by unknown sector bound nonlinearities and input saturation. An observer is designed to provide an estimate of the unknown nonlinearity. Such an estimate is used by an additional compensation loop with the purpose of mitigating the effect of the nonlinearity and enlarge the basin of attraction of the system. Stability conditions are given as a set of matrix inequalities and quadratic Lyapunov functions are exploited. An optimal compensator design algorithm based on semidefinite programming is proposed.

ABSTRACT. Firing rate models are dynamical systems abundantly exploited in applied and theoretical neuroscience to describe local cortical dynamics from real neuronal strata. The choice of the model is based upon its biological plausibility and similarity to the Hopfield model, for which theoretical results on convergence to desired state configurations and stability are well known. However, the literature seems to lack an exhaustive analytical treatment of the firing rate model, from the choice of suitable activation functions to the design of synaptic matrices, which encode how neurons in the network interact. This extended abstract provides a thorough, self-contained account on how to design firing rate models with guarantees on pattern stability.

ABSTRACT. The aim of this contribution is to illustrate the recent results obtained within the framework of the PNRR project FAIR at the University of Catania. The extension of neural networks by considering a hypercomplex topology (QNNs), and its optimization, and models for the neuron dynamics based on nonlinear matrix differential equations, referred to as hyper-neurons, have provided innovative results that are the basis of a novel paradigm of high-dimensional neural networks.

ABSTRACT. In this paper, we formulate and solve a residual generation problem that incorporates elements from both set-based estimation and unknown input observers. The objective is to determine whether the unknown input variables (e.g., possible faults) lie in a generic convex set of the input (fault) space, without needing to estimate the exact value of these variables.

ABSTRACT. The paper will show an application of the Hardware In the Loop (HIL) approach realized to verify the stability properties of the implemented fractional order PIλDμ controller. HIL simulation is a form of real-time simulation used for testing controller designs. It allows to evaluate how the designed controller reacts in real-time to realistic virtual stimuli and helps validate plant model and fine tune the controller. The HIL approach can help, in a realist way, in verifying the performance of the fractional controller, in terms of approximation, CPU speed, quantization and DAC-ACD effects, and at the same time allows to test its stability versus plant parameters variations.

ABSTRACT. With the advancements in the radiation delivery techniques and modern Linac systems, the need for better quality control devices also arises. Different devices manufactured by companies are available at hospitals and some of these devices are found to be more accurate in one category then others. The main objective of this study to analyze the dosimetric parameters of linear accelerator was to use PTW QUICKCHECK device at radiotherapy department of BINO hospital, Bahawalpur and evaluate their performances in checking the beam uniformity and symmetry in daily QC and other required periodic QC tests. For Daily Quality Control PTW QUICKCHECK device was used daily in the morning checks for 50 days to monitor CAX, beam flatness, GT symmetry, LR symmetry, Beam Quality Factor for electron beam of 6, 9, 12, 15, 18, 22 MeV energies and photon beam of 6, 15 MV energies with 100 MU given to the QUICK CHECK device at dose rate of 300 MU/min. To ensure the stability of data monitored through QUICKCHEC repeatability and reproducibility tests were performed. PTW QUICKCHECK device can be easily setup on daily basis for daily checks. According to the results it is clear that PTW QUICKCHECK device is quite accurate with regard to symmetry measurements as all data is within tolerance range (3%). However, accuracy in flatness measurement shows uncertainties i.e for 6 MV 7.3%, for 15 MV 7.31%, for 6 MeV 16.12%, for 9 MeV 6.92%, for 12 MeV 5.92%, for 15 MeV 4.01%, for 18 MeV 4.01% and for 22 MeV 4.13% of data are within tolerance range.

ABSTRACT. The building sector plays a crucial role in the global decarbonization and energy efficiency efforts. According to the International Energy Agency’s 2022 report, this sector is responsible for 27% of global energy-related emissions. Additionally, the significant energy savings potential in buildings has made optimal control in this area a popular research topic in recent years. This project focuses on the optimization and energy management of buildings, both residential and industrial, aiming to maintain thermal comfort while reducing energy consumption costs. The proposed approach incorporates renewable energy sources, such as solar panels, and includes battery energy storage and the use of the building structure as thermal storage. Since the thermal comfort can strongly depend on the thermal behavior of the building, a first analysis will be devoted to design the building such that the variation of the internal temperature is significantly small with respect to the daily variation of weather conditions. The building is considered as a thermal storage unit, where the building behaves as a low-pass filter, whose characteristics depend on its thermal characteristics, such as the thermal transmittance and capacity. In addition, a second approach involves controlling the building’s temperature according to the user's preferences by adjusting the output air temperature and scheduling the operation times of a HVAC system, exploiting the thermal storage characteristics of the building to reduce the economic and environmental cost of the use of energy. A grey-box model is suggested to estimate the building's thermal parameters through a physical thermal model. In this context, a physics-informed neural network can be applied to speed up the computing time. The optimization problem can be framed as a constrained optimization problem or a multi-objective optimization problem, depending on the chosen scenarios. This study highlights the potential for significant energy savings and cost reductions through optimized energy management and control in the building sector, contributing to broader decarbonization efforts.

ABSTRACT. Predicting the response to Trans-Arterial Chemoembolization (TACE) is crucial to personalize the treatment of hepatocellular carcinoma, improving therapeutic efficacy and prognosis and reducing side effects. This study has the aim to design and develop an autonomous intelligent system to predict the response to TACE using advanced machine learning techniques. Pre-procedure CT and clinical data from 105 patients were employed and time-to-progression treatment was used to discriminate if a patient is responsive or non-responsive. Different machine learning models were tested with a narrow selection of both clinical and radiomics features as input to discover the most effective way to assist doctors in their clinical practice.

ABSTRACT. Colorectal cancer is the third most frequently diagnosed cancer globally. Colonoscopy is crucial for screening, aiming to detect and treat malignant or pre-malignant polyps. However, the accuracy of this procedure is influenced by factors such as physician expertise, lighting variations, artifacts and differences in imaging modalities. This study conducts a comprehensive literature review on automatic polyp segmentation and detection methodologies, emphasizing unsupervised methods.

ABSTRACT. Autonomous decision-making has always been one of the primary goals to pursue as concerns mobile robots. Researchers of this field have recently turned their attention to Deep Reinforcement Learning (DRL), achieving state-of-the-art results and often outperforming classical control methods in both robustness and optimality. One aspect that is often neglected when it comes to robots' long-term independence is the energy-awareness, namely the ability of autonomously managing the charge state as a function of current and future needs. We present a Double Deep Q Network (DDQN) architecture for high level control of a mobile robot involved in site servicing tasks on a graph map. The proposed DDQN reward scheme is designed specifically to meet an energy-aware behaviour, by penalizing both extremely low levels of battery charge as well as unnecessary recharges. The model is numerically simulated on a graph scenario constituted of failure/target locations, charging station and tool change nodes. Results show that the trained agent always manages to succeed in reaching the destination without ever incurring in a complete discharge, as it promptly performs temporary stops at charging locations whenever needed.

ABSTRACT. This paper investigates the use of Motor Evoked Potentials (MEPs) and a Digital Twin (DT) approach to predict motor disability progression in patients with Multiple Sclerosis (MS). MEPs, recorded from the abductor pollicis brevis (APB) and abductor hallucis (AH) muscles, provide insights into neural conduction and motor cortex functionality. The DT is proposed as a new paradigm able to implement an individualized innovative management of MS through artificial intelligence models, copying with the complexity of this multidimensional disease. Generative Artificial Intelligence (GAI) driven DT for data management, analysis and modeling can be a promising paradigm for precision medicine. In this paper DL GAI models have been implemented as Vision Transformer (ViT) for classification and a Class-Informed Variational Autoencoder (CI-VAE) for data augmentation considering a longitudinal dataset to predict disability progression based on changes in the Expanded Disability Status Scale (EDSS).

ABSTRACT. Water pollution and global warming are major issues affecting our society. Microalgae offer a promising solution by purifying water through reproduction and photosynthesis and reducing global warming by consuming CO2. For this process to be effective, it must be maintained in optimal conditions. Therefore, a model-based control algorithm is being studied to ensure optimality. The project's goal is to develop a framework to enhance process efficiency, making it cost-effective and sustainable.

ABSTRACT. A pricing mechanism for the charging schedule problem of Electric Vehicles (EVs) that takes into account the requests of EVs is proposed. Conventional price schemes adopted in the literature are typically affected by total energy demand and the time of use (TOU) tariff. These models do not capture the fact that, in real-world booking systems, by increasing the request in the presence of fixed limited resources, the price increases. In this paper, we propose a price function that, in addition to considering the effect of total energy demand and TOU tariff, it also accounts for satisfied and dissatisfied EVs due to fixed finite resources. The problem is modeled by using a mixed-integer linear cooperative structure by including the presence of infrastructural limitations such as the number of chargers and maximum demand from the grid. A corresponding optimization problem is defined whose cost function aims to reach a compromise between the demand from the grid during peak hours and the level of dissatisfaction with electric vehicles. Simulation results show that the proposed pricing mechanism, which also accounts for resource limitation, does not have adverse effects on energy peak shaving and allows a realistic balance for EV charging satisfaction.

ABSTRACT. Using a geo-referenced agent-based model and game theory for the decision-making we investigate the climate awareness and attitudes toward sustainability in the Italian city of Siena, characterized by a population of predominantly selfish individuals and in which there are also herding and anti-herding mechanisms. We claim that on the one hand, the conformation and geographical position of the city make it naturally sustainable, but at the same time its urban structure and connection networks seem to limit the spread of cooperative actions toward environmental protection.

ABSTRACT. The ability to achieve precise and smooth trajectory tracking is crucial for ensuring the successful execution of various tasks involving robotic manipulators. State-of-the-art techniques require accurate mathematical models of the robot dynamics, and robustness to model uncertainties is achieved by relying on precise bounds on the model mismatch. In this paper, we propose a novel adaptive robust feedback linearization scheme able to compensate for model uncertainties without any a-priori knowledge on them, and we provide a theoretical proof of convergence under mild assumptions. We evaluate the method on a simulated RR robot. First, we consider a nominal model with known model mismatch, which allows us to compare our strategy with state-of-the-art uncertainty-aware methods. Second, we implement the proposed control law in combination with a learned model, for which uncertainty bounds are not available. Results show that our method leads to performance comparable to uncertainty-aware methods while requiring less prior knowledge.

ABSTRACT. Designing modern cancer therapies requires quantitative approaches to determine the personalized dose for achieving the optimal biological effects in a patient. We propose a quantitative systems biology methodology, in which a mathematical model is calibrated on the outcomes of a large-scale gastric cancer cohort that received a combination treatment. The aim of this approach is to unveil the relationships between the dose-exposure and the dose-response to improve disease control and patient survival.

ABSTRACT. This paper proposes a new active disturbance rejection control (ADRC) for induction motor (IM) drives. In particular, it improved the classic ADRC based on Extended State Observer (ESO), adopting, in its place, a high-gain Unknown Input Observer (UIO) with a driving term that is a function of the tracking error. This approach allows total robustness versus inverter nonlinearity effects and voltage drops as well as versus wrong knowledge of the inertia of the system, as explained in the paper. The proposed control technique has been experimentally verified on a suitably devised test set-up. Finally, the proposed ADRC based on the UIO has been verified in both numerical simulation and experimentally with the classic field-oriented control (FOC), the input-output feedback linearization (FLC), and an advanced ADRC based on ESO with a sliding mode (SM) action.

ABSTRACT. Ensuring secure spacing between vehicles is vital for road safety, efficient traffic flow, and system stability in autonomous driving. While traditional cooperative platooning approach, relying on centralized coordination exploiting wireless network, faces practical implementation challenges due to communication constraints and diverse driving behaviors, this work introduces a scalable non-cooperative multi-agent platooning strategy based on Deep Reinforcement Learning, leveraging on decentralized decision-making principles.

ABSTRACT. Agriculture faces new challenges in meeting the demands of a rapidly expanding global population, the constant reduction of the labor force, and significant climate changes. The integration of robotics into agricultural practices presents a promising solution to these critical issues, potentially enhancing efficiency, productivity, and sustainability in the sector.

Our research focuses on autonomous navigation in rows of orchards based on RGB-D data, a process that can be exploited in several more complex tasks such as plant monitoring, weed removal, pesticide application, precision planting, and automated harvesting. This task is particularly challenging in perception and control due to the environment's variability and irregularity. For this reason, our proposed method relies on local information, such as the disparity map and its depth representation, with minimal prior path planning, making this autonomous navigation module adaptable to any row-based cultivation structure, such as vineyards, apple orchards, or hydroponic strawberry systems.

The core algorithm employs a finite state machine that alternates between two primary states: in-row navigation and out-row navigation. The state transitions are based on a perception routine delineating accessible and obstructed regions in the rover's path. Additionally, a collision detection algorithm runs in the background to ensure safety.

The in-row navigation implements proportional control in response to the square of the horizontal deviation between the camera center and the in-row endpoint. The endpoint is determined as the center of mass of the most significant area beyond a pre-defined depth value acquired by the RGB-D camera.

The out-row navigation, instead, is based on a circular trajectory, which is dynamically adjusted according to the presence of some entity in the areas. In particular, the desired linear velocity is reduced when one is detected in the central area proportionally with respect to the square of its distance; similarly, the angular one is reduced according to entities in the turning side area.

After extensive testing in a simulated environment, we have successfully demonstrated the effectiveness of our system in actual orchard rows. The results in both settings were promising, showcasing the potential of our approach. Moving forward, we aim to conduct further experiments in diverse scenarios to validate the versatility and robustness of our method.

ABSTRACT. In the context of artificial intelligence classification problems, counterfactuals represent the minimum change that should occur in an instance to observe a diverse outcome from the classifier. Counterfactuals have been widely applied to classification problems where classes are typically associated with risks, for instance with risk of developing a given disease or not, thus classes representing safe and unsafe sets. From a theoretical perspective, counterfactuals suggest what should have been different in an instance to change with minimum effort the outcome of the AI algorithm. Indeed, although providing more insights with respect to the explainability of the decision of the algorithm, the AI-driven counterfactuals neglect the underlying dynamics of the considered system, consequently providing merely conceptual decisions, representing "virtual" conditions for the instances undergoing the change of class. In this work we propose a control system theoretical formulation for counterfactuals, with the aim of assessing a physics-informed approach suitable to account for the underlying mechanisms driving the change of class. The control framework is derived by means of an optimal control problem, aimed at computing the minimum control law steering the system to the safe set. This problem is then cast in an infinite-dimensional problem in the space of the measures and subsequently solved by means of the moment-sum of squares (moment-SOS) hierarchy through a sequence of convex relations in the space of the moments, with the aim of deriving a general methodology suitable to be exploited for both linear and nonlinear systems. This workis prodromal to the integration of control and AI method to derive physics-informed personalized minimum recommendations for disease prevention.

ABSTRACT. In this work, we present an experimental study aimed at identifying the governing equations of a nonlinear circuit from an experimental data. The procedure is based on the sparse identification method introduced in S. L. Brunton et.al , and the ultimate goal of this work is to provide an experimental validation of this procedure. The identification process involves acquiring the state variables from the circuit and their derivatives. These derivatives can either be directly derived from the circuit or computed numerically. To model the circuit's behavior, we construct a library of potential nonlinear terms. This library includes a range of candidate terms, encompassing both linear and nonlinear functions of the circuit's state variables. By leveraging sparse regression techniques, we iteratively refine this library to identify the minimal set of terms necessary to describe the system dynamics. This process includes the identification of unknown factors, which are the sparse coefficient vectors corresponding to each individual column. These individual sparse coefficient vectors are then used to construct the sparse matrix. The iterative refinement process ensures that our model remains parsimonious while capturing the essential features of the system's behaviour. The presence of null elements in the sparse matrix allows us to eliminate spurious functions from our library. Consequently, we derive the governing equations of the system. After deriving the governing equations, we use them to reconstruct the system's behavior, resulting in the reconstructed variables and the reconstructed trajectory is then compared with the attractor from experimental data to validate our study. This comparison demonstrates the success of our study. By iteratively adjusting the model parameters to minimize the error between the actual and reconstructed data, we ensure that our model captures the underlying dynamics of Chua's circuit.

ABSTRACT. The Artificial Racing Coach is an interactive driving instructor that teaches humans to drive at the vehicle handling limits and sharpen their skill to achieve the best lap time. At the core, there is our Artificial Race Driver (ARD), a hybrid AI-optimal control based real-time motion planning and control framework that autonomously learns the vehicle manoeuvrability envelope and how to control it at the limits. Augmenting ARD with shared-control of the vehicle, auditory/visual/haptic feedbacks, and an iterative teaching scheme, we are able to teach a human drivers how to improve their performance around a racetrack starting from what the system has learnt from their driving style and skills.

ABSTRACT. The Internet of Things (IoT) is transforming our world by integrating technology into various sectors, creating smarter and more responsive environments. Challenges such as device heterogeneity, lack of universal standards, and privacy issues complicate IoT systems. With over 29 billion devices expected by 2030, autonomous configuration and management are essential. This research develops a theoretical model and testbed for an Agent-Based Automatic Management Infrastructure, enhancing IoT systems efficiency and adaptability. Our architecture uses intelligent agents to manage microservices dynamically, improving resource allocation and operational efficiency, ultimately providing a robust and flexible system.

ABSTRACT. This study primarily aims to examine the optimization and control strategies for drones within enclosed spaces (indoor logistics), an area that has not been extensively explored in the literature. To the authors best knowledge, there has not been a thorough examination of optimization and control techniques used for drones in indoor logistics in a unified study. Therefore, this wok is to fill the perceived gap within the literature.

ABSTRACT. There are diverse applications in industry, agriculture and logistics which can benefit from the use of quadrotor Uncrewed Aerial Vehicles (UAVs) with cable-suspended payloads. Their control is a challenging topic, given the nonlinear and underactuated dynamics of the UAV and the oscillations of the cargo. In this work, we propose a chattering-free sliding mode controller (SMC) for position and attitude tracking of a quadrotor with a cable-suspended payload. Firstly, the model of the UAV and the load is derived via Lagrangian approach in matrix form. Then, the system is divided into two subsystems and four sliding mode controllers are designed. The chattering effect is eliminated via approximation of the signum function with a saturation function. Moreover, our approach permits computation of the sliding surface coefficients using a simple linearization of the dynamic model and Hurwitz analysis. The stability of the closed-loop controller is proven via Lyapunov theory and its effectiveness is validated through simulations. The proposed SMC allows to track the desired references when a quadrotor of 1.1 kg performs non-aggressive maneuvers, transporting a payload of 0.2 kg with a cable of 1 m. In this scenario, our approach ensures that the oscillations of the load are always smaller than 4.6 degrees.

ABSTRACT. Abstract Autonomous systems, particularly in the realm of vehicles and robots, are at the forefront of modern technological advancements. These systems must operate efficiently in crowded and structured environments, facing significant challenges in ensuring safety and adaptability. This research aims to contribute to the field of motion planning and control for autonomous systems by combining the strengths of model-based and learning-based methodologies. The primary objective is to develop algorithms that are both safe and adaptive, addressing current limitations and pushing the boundaries of state-of-the-art technologies [1-3]. The initial phase of the project focuses on model-based methodologies, which rely on mathematical models, optimization tools, and algorithms. These approaches provide performance guarantees and ensure feasibility, though they often require significant computational resources. The research aims to optimize these methodologies, reducing their computational demands and enhancing their applicability in safety-critical scenarios. In parallel, the project explores learning-based methods, which learn from data and experiments without relying on explicit models. These methods have shown promise in non-safety-critical applications due to their efficiency and adaptability. However, their lack of interpretability presents challenges in ensuring performance guarantees. The research develops techniques to improve the interpretability and reliability of learning-based approaches, making them suitable for safety-critical applications. The core innovation of this research lies in the integration of model-based and learning-based approaches. By harnessing the strengths of both methodologies, the project aims to achieve the following: • Efficiency: Leveraging the computational advantages of learning-based methods. • Adaptability: Ensuring that the systems can handle diverse and complex scenarios. • Performance Guarantees: Providing safety and reliability through model-based methods.

ABSTRACT. Transporting deformable materials like fabrics, composites or cables in robotics is challenging due to their tendency to slide, stretch, and deform. Common approaches in literature use force sensors or depth images and neural networks (NN) to adjust the fabric's state, each having drawbacks like force control effects and NN reliability. The proposed method estimates deformability constraints with geometric parameters to plan a collaborative robot's trajectory, ensuring these constraints. The robot tracks human movement safely using a near-time-optimal control strategy.

ABSTRACT. Social robot navigation has evolved significantly, focusing on safe and efficient human-robot interaction through rapid decision-making and human-like navigation protocols. Current methods, such as social force models, may lack robust deadlock-free guarantees and struggle in complex scenarios. We propose a novel approach that combines potential fields with nonlinear opinion dynamics, ensuring reliable, collision-free navigation in dynamic human environments. The use of potential fields, enhanced by an oval-shaped vortex field around humans, allows to reduce trajectory curvature and to respect personal space, while the integration of opinion dynamics helps the robot to choose the direction of rotation around obstacles. Simulation and experimental results demonstrate the robot's ability to reach its goal while efficiently avoiding humans, with dynamic opinion adjustments guiding smooth manoeuvres. This approach offers a promising solution for socially-aware robot navigation, enhancing human-robot coexistence.

ABSTRACT. In this paper, we propose a new piecewise constant feedback for string-stability of a platoon under sampled and quantized measurements. The design is based on a mesoscopic approach and is carried out over the sampled-data model associated to each vehicle. The proposed feedback ensures string-stability in the practical sense independently of the effect of sampling and quantization. Simulations show the effectiveness of the results.

ABSTRACT. With the rise in frequency of environmental diseases like wildfires and landslides, protection and risk management have become critical challenges for assuring both the safety of human populations and the sustainability of ecosystems. In this direction, a suite of functionalities and solutions have been developed to support emergency operators in all phases of emergency management, from risk estimation to enviromental monitoring, early detection and response. The present paper provides an overview of the main algorithms that consistute the buidling blocks of the enviromental monitoring system developed in the ARIES project, founded by ESA, through a combination of dynamical systems, digital twins and convolutional neural networks, also reporting on its early validation activities.

ABSTRACT. The estimation accuracy in Extended Kalman Filters is highly dependent on selecting appropriate process and noise covariance matrices. Monte Carlo optimization and numerical gradient descent techniques are employed to accurately determine the optimal structure and value for both the covariance matrices, achieving enhanced performance. This approach will be tested on a localization problem utilizing UWB, IMU, and odometric sensors, using experimental data for validation.

ABSTRACT. High penetration of renewable energy sources (RES) adds further uncertainty to the power systems, which are complex and nonlinear as well. Reinforcement learning (RL) approaches, due to their good performance with stochastic nonlinear systems, recently attracted interest in the control of power systems, including voltage regulation. However, ensuring the safety of these controllers is challenging because of potential instabilities. This work proposes a decentralized safe RL framework for voltage control using inverter-based DERs, guaranteeing exponential stability with specific Lipschitz constraints. Optimizing these bounds expands the search space for neural network controllers. The framework trains the neural network of each busbar using local measurements, eliminating the need for communication links. The method is tested on IEEE 15 and 33 busbar networks.

ABSTRACT. Contracting dynamical systems exhibit remarkable convergence and robustness properties. In recent years, these properties have made contraction theory a valuable tool for developing robust controllers and advanced convergence analysis techniques. In this work, we leverage contraction theory alongside neural networks to develop a non-linear feedback controller capable of performing trajectory-tracking tasks effectively.

ABSTRACT. A novel optimal LQR state-feedback control law is proposed for energy harvesting maximization in regenerative suspension systems where an actively governed electromechanical actuator is used in place of the viscous damper. A special LQR cost function is considered that directly maximizes the electrical power generated by the electromechanical actuator. Other conflcting control objectives, such as ride comfort and road handling, may be considered along with the energy harvesting objective in the proposed control setup, allowing one to directly trade-off among them depending on the application. Specifially, as an example, a condition for trading-off between energy harvesting and ride comfort is added to the optimization problem via forcing a bound on the so called Ride Index. The proposed control law is fially contrasted with other two control strategies usually considered in the literature for energy harvesting applications and is compared in simulative studies via MATLAB and Simulink on a quarter-car model.

ABSTRACT. Electrolyte-gated carbon nanotube field-effect transistors (EG-CNTFETs) are promising platforms for biosensing applications, but struggle with instability issues. This contribution demonstrates that incorporating a lipophilic membrane on the semiconducting channel significantly reduces stabilization time to 34 minutes, compared to an hour for state-of-the-art devices. This enhancement also shows with the on-off ratio increasing from 9.76 ± 4.16 to 24.55 ± 20.99 in one hour. In contrast, devices without the membrane exhibit a decrease in the on-off ratio, dropping from 88.7 ± 29.9 to 61.1 ± 23.9. Moreover, the sensor performance was extended to up to 12 hours and 5 repeated uses. As proof-of-concept, encapsulated devices were functionalized with an ion-selective membrane showing a linear detection range for ammonium ion from 0.01 mM to 100 mM, with a coefficient of determination of 94.71% and a sensitivity of 0.143 A/decade. These advancements mark significant progress towards stable and functional EG-CNTFET-based biosensors for diverse food, biomedical, and environmental applications. Furthermore, the potential to embed these transistors in a robotic system could enable fully automated analysis, from sample collection through a microfluidics system to the sensing component, and finally to a readout system that delivers the analytical results.

ABSTRACT. Batch manufacturing process remains the primary method for industrial pro- duction. The goals in a batch process are: (I) Reliability, by minimizing op- erational risks; and (II) Efficiency, providing high production standards while reducing the production cost. However, it is essential to consider and closely monitor the significant drawbacks associated with a batch manufacturing pro- cess: (I) Complex and dynamic systems that are highly susceptible to disruptions or deviations in each phase; (II) Quality control issues within a batch can lead to entire batches being rejected; and (III) Inconsistent product quality between batches can lead to inconsistent product quality if the manufacturing process is not optimized.

ABSTRACT. Automating and optimizing wood processing have been of significant importance to improve efficiency in sawmills. This study presents a quality-driven constructive heuristic approach for solving the live sawing optimization problem. This deals with cutting rectangular boards from cylindrical logs with circular cross sections. The heuristic is adaptive regarding the existence of defects inside the logs that can be detectable after each cut. Thus, the cutting pattern is updated after each cut with new scanned information about internal defects in the log. The problem is a two-dimensional strip bin-packing problem with the objective of orthogonal packing of a given set of rectangular items into a set of strips inside a circle to maximize the sawing yield. We apply a combination of sorting, placement, and searching policies, along with an online strategy to build a feasible cutting that enables guillotine cuts and quality consideration for wood sawing. Our approach outperforms the current state of the art in improving sawing efficiency.

ABSTRACT. Large-scale multi-agent systems are increasingly relevant in various aspects of society. The operation of these systems requires advances in multi-agent distributed optimization algorithms that can handle uncertain environments. This work presents the problem of developing distributed robust algorithms suitable for solving convex constraint-coupled multi-agent uncertain problems, where the uncertainty directly affects the coupling constraints. To test the proposed algorithm, we apply it to a multi-microgrid aggregation problem to provide ancillary services to the Grid, a relevant topic in decarbonisation and energy security.

ABSTRACT. Literature shows that direct methods are the most popular approach for solving optimal control problems due to their robustness and effective constraint handling. Conversely, indirect methods, though more challenging to implement, offer greater accuracy by deriving necessary optimality conditions via Pontryagin's Maximum Principle and solving two-point boundary value problems. These methods provide deeper insights into the problem structure, crucial for theoretical analysis and efficient algorithms development. This study demonstrates that it is possible to construct numerical approximations that leverage both direct and indirect methods, inheriting the advantages and properties of each.

ABSTRACT. We propose a novel continuous-time algorithm for inequality-constrained strongly convex optimization inspired by proportional-integral control. Compared to the popular primal-dual gradient dynamics, our approach simplifies the proof of exponential convergence. Moreover, through several examples, we show that the proposed algorithm may converge faster than primal-dual gradient dynamics.

ABSTRACT. Trading off exploration and exploitation is a crucial task in global (or black-box) optimization, to balance the search for better local optimizers with the refinement of already-found ones. Often, such a trade-off is not easily controlled by the user, as it depends non-trivially on the tuning parameters of the selected algorithm. A new concept is proposed here, where the share of exploitation moves over the total number of iterations is regulated by a feedback control law, to achieve a user-defined set-point. This concept is applied to the recently proposed Set Membership Global Optimization (SMGO) technique, resulting in a modified algorithm. Together with computational improvements the resulting approach is extensively tested and compared with other methods. The statistical tests indicate that the new algorithm has better iteration-based optimization performance than the original one, at the same time shortening the computational times by around one order of magnitude. This extended abstract is based on a recently presented conference paper.

ABSTRACT. This work presents a novel robust variable-horizon model predictive control scheme designed to intercept a target moving along a known trajectory, in finite time. Linear discrete-time systems affected by bounded process disturbances are considered and a tube-based MPC approach is adopted. The main contribution is an adaptive mechanism for choosing the terminal constraint set sequence in the MPC optimization problem. This mechanism is designed to ensure recursive feasibility while promoting minimization of the final distance to the target. Finite-time convergence of the proposed control scheme is proven. In order to evaluate its effectiveness, the designed control law is tested through numerical simulations, including a case study involving orbital rendezvous of a satellite with a tumbling object. The results indicate a significant reduction in conservatism compared to existing state-of-the-art methods using a fixed terminal set sequence.

ABSTRACT. Quadratic programming (QP) solvers that join effectiveness with a simple implementation are becoming essential in the field of optimal control, specifically when dealing with real-time applications with strict timing constraints and limited computational resources. To address this need, we present a novel high-performance QP solution method based on pseudo-transient continuation (PTC). PTC is a numerical technique that transforms multivariate nonlinear equations into autonomous systems that converge to the solution sought. In our approach, we recast the general QP Karush-Kuhn-Tucker (KKT) conditions into a system of equations and employ PTC to solve the latter to attain the optimal solution. Importantly, we provide theoretical guarantees demonstrating the global convergence of our PTC-based solver to the optimal solution of any given QP. To showcase the effectiveness of PTC, we employ it within the domain of Model Predictive Control (MPC). Specifically, numerical simulations are carried out on the MPC control of a quadrotor – a demanding dynamical system – highlighting excellent results in accurately executing the control task and ensuring lower computational times compared to conventional QP solvers.

ABSTRACT. Integrating renewable energy sources into the power grid introduces significant uncertainty, making accurate wind power forecasting crucial. This study compares two forecasting methods: parametric Beta regression with natural spline preconditioning and non-parametric Quantile Regression Forests (QRFs). Using SCADA data from the Penmanshiel wind farm and wind forecasts from the Copernicus project, models were evaluated for a 6-hour ahead forecasting horizon. Beta regression effectively captured complex power distribution patterns influenced by varying wind speeds, while QRFs, despite providing valid probabilistic forecasts, exhibited higher variance and required meticulous hyperparameter tuning. The findings underscore the necessity of flexible modeling for accurate wind power forecasts.

ABSTRACT. Given the widespread attention to individual thermal comfort, coupled with significant energy-saving potential inherent in energy management systems for optimizing indoor environments, this paper aims to introduce advanced "Humans-in-the-building" control techniques to redefine the paradigm of indoor temperature design. Firstly, we innovatively redefine the role of individuals in the control loop, establishing a model for users' thermal comfort and constructing discomfort signals based on individual preferences. Unlike traditional temperature-centric approaches, "thermal comfort control" prioritizes personalized comfort. Then, considering the diversity among users, we propose a novel method to determine the optimal indoor temperature range, thus minimizing discomfort for various users and reducing building energy consumption. Finally, the efficacy of the "thermal comfort control" approach is substantiated through simulations conducted using Matlab.

ABSTRACT. This paper proposes a novel approach for the optimal scheduling and control of a microgrid with an electrolyzer and fuel cell systems, both of proton exchange membrane (PEM) technology. It is based on a hierarchical procedure constituted of two levels of optimization: a higher level based on an economic optimization for the optimal scheduling of different components of the microgrid, and a lower level for the real-time control of the hydrogen systems: PEM electrolyzer (PEMEZ) and PEM fuel cell (PEMFC). The flexible operation imposed by the higher level leads to a violation of the limits designed by the manufacturer of the hydrogen components, specifically when switching from one power level to another. The current proportional-integral (PI) controllers integrated into those systems cannot handle this issue, which provokes a premature aging phenomenon of the materials and leads to poor performance of the systems. In the present paper, at the lower level, a reference governor (RG) real-time control approach has been added on top of the PI controller to ensure the respect of the operating limits and guarantee better performances. The focus has been given to the stack's temperature in both the electrolyzer and fuel cell systems as the control objective because of its direct influence on the material's durability and, by extension, on the efficiency. The bi-level optimization and control architecture has been applied and validated through simulations using data from a real-world case study, specifically the Savona Campus Smart Polygeneration Microgrid in Italy. The results showed a significant reduction in the overshoots of the stack's temperature compared to the PI controller.

ABSTRACT. This study investigates deep offshore, pumping Airborne Wind Energy systems, focusing on the kite-platform interaction. The considered system includes a 360 m2 soft-wing kite, connected by a tether to a winch installed on a 10-meter-deep spar with four mooring lines. Wind power is converted into electricity with a feedback controlled periodic trajectory of the kite and corresponding reeling motion of the tether. An analysis of the mutual influence between the platform and the kite dynamics, with different wave regimes, reveals a rather small sensitivity of the flight pattern to the platform oscillations; on the other hand, the frequency of tether force oscillations can be close to the platform resonance peaks, resulting in possible increased fatigue loads and damage of the floating and submerged components. A control design procedure is then proposed to avoid this problem, acting on the kite path planner. Simulation results confirm the effectiveness of the approach.

ABSTRACT. As the advent of distributed energy resource technologies has increased the independence of grid actors from central power providers, the problem of efficiently controlling the overall operational apparatus has not yet thoroughly been solved [1]. The problem of distributed Generalized Nash Equilibrium (GNE) computation in multi-agent systems has recently received significant attention from game theoretical perspectives [2]. This abstract summarizes [3], which extends [4], where the community members have private cost functions coupled with each other through the monetary price at which each member sells its energy. As the main technical contribution, we provide sufficient conditions on the private cost and energy pricing functions to ensure the strong monotonicity of the overall pseudo-gradient mapping, which is in turn a sufficient condition to guarantee the convergence of the considered ADMM-type algorithm.

ABSTRACT. The quest for energy efficiency and consumption control is fundamental from an environmental perspective. Fifty percent of building energy consumption is used in Heating, Ventilation, and Air-Conditioning (HVAC) systems. Hence, effective HVAC control techniques for energy consumption minimization and thermal comfort guarantee have attracted the attention of researchers. To simultaneously maximize comfort and minimize energy consumption, it is essential to manage the building network (district) in a way that optimally balances real-time energy usage. While existing regulations have initiated some changes, traditional penalty-based systems often fail to encourage full user compliance. Emerging researchers advocate for integrating blockchain technology, like proposing a novel system that dynamically rewards or penalizes users based on their real-time energy consumption. This approach promises enhanced security and privacy, addressing key concerns in energy management. Thus, our work introduces an innovative HVAC management system connected with the District Energy Management System (DEMS). It classifies users into Consumption Classes using a K-means clustering algorithm, determining rewards or penalties based on their energy consumption behavior, implemented through a blockchain platform to ensure transparency and security. The novelty lies in the integration of Model Predictive Control (MPC) combined with a Long Short-Term Memory (LSTM) network, which users can follow to achieve more virtuous energy consumption classes.

ABSTRACT. This work proposes a novel methodology to identify the model of a district heating system (DHS) exploiting physics-informed neural networks. The approach consists in leveraging multiple recurrent neural networks (RNNs) and in embedding the physical topology of the DHS within their interconnections. Compared to standard RNNs, this modeling technique enables faster training and higher accuracy. The developed approach paves the way for the design of a nonlinear model predictive control (NMPC) strategy, allowing to minimize production costs, enhance system efficiency and ensure compliance with operational constraints across the entire DHS.

ABSTRACT. We address economic dispatch of power generators with prohibited operating zones. The problem can be formulated as an optimization program with a quadratic cost, non-convex local operating constraints, and a scalar quadratic coupling constraint accounting for load demand and power losses. A duality-based resolution approach integrating a bisection iterative scheme is proposed to reduce computational complexity while guaranteeing finite time feasibility of the primal iterates and a cost improvement throughout iterations. Extensive simulations show that the approach outperforms state-of-the-art competitors and consistently computes feasible primal solutions with a close-to-zero optimality gap at a low computational cost.