Publications Lists     [ all BibTeX ]  

This work presents a novel general regularized distributed solution for the state estimation problem in networked systems. Resting on the graph-based representation of sensor networks and adopting a multivariate least-squares approach, the designed solution exploits the set of the available inter-sensor relative measurements and leverages a general regularization framework, whose parameter selection is shown to control the estimation procedure convergence performance. As confirmed by the numerical results, this new estimation scheme allows (i) the extension of other approaches investigated in the literature and (ii) the convergence optimization in correspondence to any (undirected) graph modeling the given sensor network.

The diffusion of the Industry 4.0 paradigm lead to the creation and collection of huge manufacturing datasets; such datasets contain for example measurements coming from physical sensors located in different equipment or even in different productive manufacturing organizations. Such large and heterogeneous datasets represent a challenge when aiming for developing data-driven approaches like Anomaly Detection or Predictive Maintenance. In this work we present a new approach for performing Anomaly Detection that is able to handle heterogeneous data coming from different equipment, work centers or production sites.

This work focuses on bearing rigidity theory, namely the branch of knowledge investigating the structural properties necessary for multi-element systems to preserve the inter-unit bearings under deformations. The contributions of this work are twofold. The first one consists in the development of a general framework for the statement of the principal definitions and properties of bearing rigidity. We show that this approach encompasses results existing in the literature, and also provides a systematic approach for studying bearing rigidity on any differential manifold in SE(3)^n, where n is the number of agents.The second contribution is the derivation of a general form of the rigidity matrix, a central construct in the study of rigidity theory. We provide a necessary and sufficient condition for the infinitesimal rigidity of a bearing framework as a property of the rank of the rigidity matrix. Finally, we present two examples of multi-agent systems not encountered in the literature and we study their rigidity properties using the developed methods

Recent results show that features of adversarially trained networks for classification, in addition to being robust, enable desirable properties such as invertibility. The latter property may seem counter-intuitive as it is widely accepted by the community that classification models should only capture the minimal information (features) required for the task. Motivated by this discrepancy, we investigate the dual relationship between Adversarial Training and Information Theory. We show that the Adversarial Training can improve linear transferability to new tasks, from which arises a new trade-off between transferability of representations and accuracy on the source task. We validate our results employing robust networks trained on CIFAR-10, CIFAR-100 and ImageNet on several datasets. Moreover, we show that Adversarial Training reduces Fisher information of representations about the input and of the weights about the task, and we provide a theoretical argument which explains the invertibility of deterministic networks without violating the principle of minimality. Finally, we leverage our theoretical insights to remarkably improve the quality of reconstructed images through inversion.

In this work we focus on the problem of minimizing the sum of convex cost functions in a distributed fashion over a peer-to-peer network. In particular, we are interested in the case in which communications between nodes are prone to failures and the agents are not synchronized among themselves. We address the problem proposing a modified version of the relaxed ADMM, which corresponds to the Peaceman-Rachford splitting method applied to the dual. By exploiting results from operator theory, we are able to prove the almost sure convergence of the proposed algorithm under general assumptions on the distribution of communication loss and node activation events. By further assuming the cost functions to be strongly convex, we prove the linear convergence of the algorithm in mean square in a neighborhood of the optimal solution, and provide an upper bound to the convergence rate. Finally, we present numerical results testing the proposed method in different scenarios.

The potentialities of In-Orbit Servicing (IOS) to extend the operational life of satellites and the need to implement Active Debris Removal (ADR) to effectively tackle the space debris problem are well known among the space community. Research on technical solutions to enable this class of missions is thriving, also pushed by the development of new generation sensors and control systems. Among private companies, space agencies and universities, the European Space Agency (ESA) has been developing technologies in this field for decades. Several solutions have been proposed over the years to safely capture orbital objects, the majority relying on robotic systems. A promising option is the employment of an autonomous spacecraft (chaser) equipped with a highly dexterous robotic arm able to perform the berthing with a resident space object. This operation poses complex technical challenges both during the approach phase and after contact. In this respect, the design of an effective, reliable, and robust Guidance, Navigation and Control (GNC) system, for which several algorithmic architectures and hardware configurations are possible, plays a key role to ensure safe mission execution. This work presents the outcomes of a research activity performed by a consortium of universities under contract with ESA with the goal to develop the navigation and control subsystems of a GNC system for controlling a chaser equipped with a redundant manipulator. Both the final approach until capture and the target stabilization phase after capture are considered in the study. The proposed solution aims at the implementation of a combined control strategy. Robust control methods are adopted to design control laws for the uncertain, nonlinear dynamics of the chaser and of the complete chaser–target stack after capture. Visual–based solutions, i.e., relying on active/passive electro– optical sensors, are selected for relative navigation. A complete sensor suite for relative and absolute navigation is part of the GNC system, including transducers for robot joint measurements. To properly validate the proposed solutions, a complete numerical simulator has been developed. This software tool allows to thoroughly assess the system performance, accounting for all the relevant external disturbances and error sources. A realistic synthetic image generator is also used for relative navigation performance assessment. This paper presents the design solutions and the results of preliminary numerical testing, considering three mission scenarios to prove the flexibility of the solution and its applicability to a wide range of operational cases.

Deep Learning approaches have revolutionized in the past decade the field of Computer Vision and, as a consequence, they are having a major impact in Industry 4.0 applications like automatic defect classification. Nevertheless, additional data, beside the image/video itself, is typically never exploited in a defect classification module: this aspect, given the abundance of data in data-intensive manufacturing environments (like semiconductor manufacturing) represents a missed opportunity. In this work we present a use case related to Scanning Electron Microscope (SEM) images where we exploit a  Bayesian approach to improve defect classification. We validate our approach on a real-world case study and by employing modern Deep Learning architectures for classification.

This work proposes a formation control law for multi-agent systems whose components are heterogeneous in terms of actuation capabilities, but at the same time are all able to retrieve bearing information w.r.t. some neighbors in the group. The designed controller exploits the results of the bearing rigidity theory deriving from the modeling of heterogeneous formations as generalized frameworks. The outlined solution is compared with a leader-follower combination of existing rigidity based homogeneous formation controllers in order to highlight the easy tuning, the flexibility w.r.t. the formation composition, and the increased efficiency of the new proposed control approach. A sufficient condition ensuring the convergence of the designed controller is also given.

Visual inspection has recently gained increasing importance in the manufacturing industry and is often addressed by means of learning methodologies applied to data obtained from specific lighting and camera system setups. The industrial scenario becomes particularly challenging when the inspection regards reflective objects, which may affect both the data acquisition and the classification decision process, thus limiting the overall performance. In this context, we observe that the dynamics of the reflected light is the key aspect to characterize these surfaces and needs to be accurately exploited to improve the performances of the learning algorithms. To this aim, we propose a combined model-based and data-driven approach designed to detect defects on the reflective surfaces of industrial products, captured as video sequences under coaxial structured illumination. Specifically, a tunable spatial-temporal descriptor of the evolution of the reflected light (Dynamic Evolution of the Light, DEL) is designed and employed within a Hybrid Learning (HL) framework, where the learning process of a Convolutional Neural Network (CNN) is driven by the model-based descriptor. This approach is also extended by adopting the similar in nature descriptor Dynamic Image. The proposed HL solutions are validated against a whole spectrum of state-of-the-art learning procedures and different descriptors. Experiments run on a dataset coming from an actual industrial scenario confirm the ability of DEL to accurately characterize reflective surfaces and the validity of the HL method, which shows remarkably better performance in fault detection even with respect to modern 3D- CNNs with comparable computational effort.

Particle accelerators are complex systems composed of multiple subsystems that must work together to produce high quality beams employed for physics experiments. A fault or an anomalous behaviour in one of such subsystems can lead to expensive downtime for the whole facility. Thus, it is of paramount importance to be able to promptly detect anomalies.Given the vast amount of streaming data generated by accelerator field sensors, Machine Learning (ML)-based tools are promising candidates for efficient monitoring of such systems: an approach based on unsupervised ML techniques exploiting the data from a Radio Frequency tuning system is here proposed. Feature importance is exploited to guide the definition of the optimal windowing for feature extraction. The proposed approach is here validated on real-world data related to the ALPI accelerator at Legnaro National Laboratories in Italy.

The transportation of large payloads can be made possible with Multi-Robot Systems(MRS) implementing cooperative strategies. In this work, we focus on the coordinated MRS trajectory planning task exploiting a Model Predictive Control (MPC) framework addressing both the actingrobots and the transported load. In this context, the main challenge is the possible occurrence of a temporary mismatch among agents’ actions with consequent formation errors that can cause severe damage to the carried load. To mitigate this risk, the coordination scheme may leverage a leader–follower approach, in which a hierarchical strategy is in place to trade-off between the task accomplishment and the dynamics and environment constraints. Nonetheless, particularly in narrow spaces or cluttered environments, the leader’s optimal choice may lead to trajectories that are infeasible for the follower and the load. To this aim, we propose a feasibility-aware leader–follower strategy, where the leader computes a reference trajectory, and the follower accounts for its own and the load constraints; moreover, the follower is able to communicate the trajectory infeasibility to the leader, which reacts by temporarily switching to a conservative policy. The consistent MRS co-design is allowed by the MPC formulation, for both the leader and the follower: here, the prediction capability of MPC is key to guarantee a correct and efficient execution of the leader–follower coordinated action. The approach is formally stated and discussed, and a numerical campaign is conducted to validate and assess the proposed scheme, with respect to different scenarios with growing complexity.

The functional architecture of the resting brain, as measured with the blood oxygenation level-dependent functional connectivity (BOLD-FC), is slightly modified during task performance. In previous work, we reported behaviorally relevant BOLD-FC modulations between visual and dorsal attention regions when subjects performed a visuospatial attention task as compared to central fixation (Spadone et al., 2015).

Here we use magnetoencephalography (MEG) in the same group of subjects to identify the electrophysiological correlates of the BOLD-FC modulation found in our previous work. While BOLD-FC topography, separately at rest and during visual attention, corresponded to neuromagnetic Band-Limited Power (BLP) correlation in the alpha and beta bands (8-30 Hz), BOLD-FC modulations evoked by performing the visual attention task (Spadone et al. 2015) did not match any specific oscillatory band BLP modulation. Conversely, following the application of an orthogonal spatial decomposition that identifies common inter-subject co-variations, we found that attention–rest BOLD-FC modulations were recapitulated by multi-spectral BLP-FC components. Notably, individual variability of alpha connectivity between Frontal Eye Fields and visual occipital regions, jointly with decreased interaction in the Visual network, correlated with visual discrimination accuracy. In summary, task-rest BOLD connectivity modulations match multi-spectral MEG BLP connectivity.

Research on connected vehicles represents a continuously evolving technological domain, fostered by the emerging Internet of Things (IoT) paradigm and the recent advances in intelligent transportation systems. Nowadays, vehicles are platforms capable of generating, receiving and automatically act based on large amount of data. In the context of assisted driving, connected vehicle technology provides real-time information about the surrounding traffic conditions. Such information is expected to improve drivers' quality of life, for example, by adopting decision making strategies according to the current parking availability status. In this context, we propose an online and adaptive scheme for parking availability mapping. Specifically, we adopt an information-seeking active sensing approach to select the incoming data, thus preserving the onboard storage and processing resources; then, we estimate the parking availability through Gaussian Process Regression. We compare the proposed algorithm with several baselines, which attain inferior performance in terms of mapping convergence speed and adaptivity capabilities; moreover, the proposed approach comes at the cost of a very small computational demand.

Active Search and Tracking for search and rescue missions or collaborative mobile robotics relies on the actuation of a sensing platform to detect and localize a target. In this paper we focus on visually detecting a radio-emitting target with an aerial robot equipped with a radio receiver and a camera. Visual-based tracking provides high accuracy, but the directionality of the sensing domain often requires long search times before detecting the target. Conversely,radio signals have larger coverage, but lower tracking accuracy. Thus, we design a Recursive Bayesian Estimation scheme that uses camera observations to refine radio measurements. To regulate the camera pose, we design an optimal controller whose cost function is built upon a probabilistic map. Theoretical results support the proposed algorithm, while numerical analyses show higher robustness and efficiency with respect to visual and radio-only baselines.

Research on wireless sensors represents a continuously evolving technological domain thanks to their high flexibility and scalability, fast and economical deployment, pervasiveness in industrial, civil and domestic contexts. However, the maintenance costs and the sensors reliability are strongly affected by the battery lifetime, which may limit their use. In this paper we consider a wireless smart camera, equipped with a low-energy radio receiver, and used to visually detect a moving radio-emitting target. To preserve the camera lifetime without sacrificing the detection capabilities, we design a probabilistic energy-aware controller to switch on/off the camera. The radio signal strength is used to predict the target detectability, via self-supervised Gaussian Process Regression combined with Recursive Bayesian Estimation. The automatic training process minimizes the human intervention, while the controller guarantees high detection accuracy and low energy consumption, as numerical and experimental results show.

Multi-modal Probabilistic Active Sensing (MMPAS) uses sensor fusion and probabilistic models to control the perception process of robotic sensing platforms. MMPAS is successfully employed in environmental exploration, collaborative mobile robotics, and target tracking, being fostered by the high performance guarantees on autonomous perception. In this context, we propose a bi-Radio-Visual PAS scheme to solve the transmitter discovery problem. Specifically, we firstly exploit the correlation between radio and visual measurements to learn a target detection model in a self-supervised manner. Then, the model is combined with antenna radiation anisotropies into a Bayesian Optimization framework that controls the platform. We show that the proposed algorithm attains an accuracy of 92%, overcoming two other probabilistic active sensing baselines.

In the world of autonomous driving high resolution maps play a fundamental role. Such maps are highly accurate representations of the environment and are essential for all the algorithms of strategy and path planning operations.

Unfortunately it is not always possible to guarantee the total reliability of these maps and therefore it is necessary to introduce a system for its validation. In this paper we introduce a framework for validating map data at run-time based on probabilistic graphical models. Results from simulations show the capabilities of the proposed approach and highlight the need to find an appropriate balance between model accuracy and complexity.