Publications

Objective: To define the prevalence of variants in collagen VI genes through a next-generation sequencing (NGS) approach in undiagnosed patients with suspected neuromuscular disease and to propose a diagnostic flowchart to assess the real pathogenicity of those variants. Methods: In the past five years, we have collected clinical and molecular information on 512 patients with neuromuscular symptoms referred to our center. To pinpoint variants in COLVI genes and corroborate their real pathogenicity, we sketched a multistep flowchart, taking into consideration the bioinformatic weight of the gene variants, their correlation with clinical manifestations and possible effects on protein stability and expression. Results: In Step I, we identified variants in COLVI-related genes in 48 patients, of which three were homozygous variants (Group 1). Then, we sorted variants according to their CADD score, clinical data and complementary studies (such as muscle and skin biopsy, study of expression of COLVI on fibroblast or muscle and muscle magnetic resonance). We finally assessed how potentially pathogenic variants (two biallelic and 12 monoallelic) destabilize COL6A1-A2-A3 subunits. Overall, 15 out of 512 patients were prioritized according to this pipeline. In seven of them, we confirmed reduced or absent immunocytochemical expression of collagen VI in cultured skin fibroblasts or in muscle tissue. Conclusions: In a real-world diagnostic scenario applied to heterogeneous neuromuscular conditions, a multistep integration of clinical and molecular data allowed the identification of about 3% of those patients harboring pathogenetic collagen VI variants.

Over the last decades, the exuberant development of next-generation sequencing has revolutionized gene discovery. These technologies have boosted the mapping of single nucleotide polymorphisms (SNPs) across the human genome, providing a complex universe of heterogeneity characterizing individuals worldwide. Fractal dimension (FD) measures the degree of geometric irregularity, quantifying how “complex”a self-similar natural phenomenon is. We compared two FD algorithms, box-counting dimension (BCD) and Higuchi’s fractal dimension (HFD), to characterize genome-wide patterns of SNPs extracted from the HapMap data set, which includes data from 1184 healthy subjects of eleven populations. In addition, we have used cluster and classification analysis to relate the genetic distances within chromosomes based on FD similarities to the geographical distances among the 11 global populations. We found that HFD outperformed BCD at both grand average clusterization analysis by the cophenetic correlation coefficient, in which the closest value to 1 represents the most accurate clustering solution (0.981 for the HFD and 0.956 for the BCD) and classification (79.0% accuracy, 61.7% sensitivity, and 96.4% specificity for the HFD with respect to 69.1% accuracy, 43.2% sensitivity, and 94.9% specificity for the BCD) of the 11 populations present in the HapMap data set. These results support the evidence that HFD is a reliable measure helpful in representing individual variations within all chromosomes and categorizing individuals and global populations.

Deductive and inductive reasoning are fundamental logical processes critical to the solution of common practical problems in daily life. We used functional magnetic resonance imaging (fMRI) to investigate the brain networks involved in Contradictory, Deductive, and Inductive judgments. The experimental paradigm was based on categorical propositions of the Aristotelian Square of Opposition (ASoO). In a full factorial design, identical sentences were combined into premise–conclusion pairs. Each sentence started with ‘every’ or ‘some’. The order of the two propositions in the pair created two types of logical operators (every→some: deductive, or some→every: inductive). The descriptive attributes of the category could be Contradictory or non-Contradictory. Imaging data was analyzed using Group Independent component analysis of fMRI Toolbox (GIFT). Connectivity of nodes within four intrinsic connectivity networks (ICNs) was sensitive to attribute manipulation (Contradiction): the anterior default mode network (aDMN), and the language and cerebellum networks were more involved in Contradictory than non-Contradictory statements, while the anterior salience network (aSN) showed the opposite pattern. Five networks were associated with logical operator manipulation. Stronger positive associations with Inductive than Deductive reasoning were observed in the dorsal and ventral parts of the aDMN, aSN, and orbitofrontal networks (OFN). A stronger negative association with deductive than inductive reasoning was observed in the executive control (ExCN) and dorsal attention (DAN) networks. Differences in the fractional amplitude of low‐frequency fluctuation of the BOLD signal in aDMN, ExCN, and OFN explained 67% of the variance of the behavioural cost of inductive relative to deductive reasoning. The results suggest that different ICNs support logical reasoning and conflict identification. Finally, the magnitude of the differences was positively correlated with behavioural cost.

The restrictions and isolation imposed by the COVID-19 pandemic have led to a greater diffusion of telepresence robots as tools enabling people to assist humans and keep the contacts with them. On the one hand, telepresence robots should implement user commands. On the other, they are supposed to behave in a social manner. This work evaluates a shared autonomy approach based on people-aware navigation for teleoperating a telepresence robot. The system considers both static and dynamic people. The results demonstrate the effectiveness of the presented shared autonomy system with respect to the traditional manual teleoperation in terms of robustness and feedback received from the participants.

This article proposes a novel shared intelligence system for brain-machine interface (BMI) teleoperated mobile robots where user’s intention and robot’s intelligence are concurrent elements equally participating in the decision process. We designed the system to rely on policies guiding the robot’s behavior according to the current situation. We hypothesized that the fusion of these policies would lead to the identification of the next, most probable, location of the robot in accordance with the user’s expectations. We asked 13 healthy subjects to evaluate the system during teleoperated navigation tasks in a crowded office environment with a keyboard (reliable interface) and with 2-class motor imagery (MI) BMI (uncertain control channel). Experimental results show that our shared intelligence system 1) allows users to efficiently teleoperate the robot in both control modalities; 2) it ensures a level of BMI navigation performances comparable to the keyboard control; 3) it actively assists BMI users in accomplishing the tasks. These results highlight the importance of investigating advanced human-machine interaction (HMI) strategies and introducing robotic intelligence to improve the performances of BMI actuated devices.

The Planning Domain Definition Language (PDDL) successfully encodes classical planning tasks by easily describing objects, actions, and states in many planning domains. PDDL also describes domains, but they include only predefined sets of actions that can solve problems in a finite set of states. Indeed, the PDDL structure disables the processing of single predicates and operators. As a consequence, they cannot be arbitrarily composed to model new domains. To overcome these limitations, we propose a domain-independent, general-purpose knowledge design and task planning system based on the combination of a PDDL generator and interpreter and a Knowledge Base. The former builds planning data structures, where every object is a PDDL token independent of its original domain. It also allows merging these objects to formulate new PDDL domains and problems, ensuring consistency and validity of generated definitions. Their resolution is based on a powerful object-based reasoning instead of an inefficient lexical-based one. The latter contains the necessary relationships and representations to allow data storing and reusability. Their combination enables the storage, interpretation, and reuse of planning data, resulting in integration between the planning process and description logic reasoning. The overall system guarantees a flexible adaptation of the computed planning domains to changing environmental conditions, agent capabilities, and assigned tasks, promoting effective sharing and reuse of domain knowledge across different systems and applications.

In shared control teleoperation, the robot assists the user in accomplishing the desired task. Rather than simply executing the user’s command, the robot attempts to integrate it with information from the environment, such as obstacle and/or goal locations, and it modifies its behavior accordingly. In this article, we propose a real-time shared control teleoperation framework based on an artificial potential field approach improved by the dynamic generation of escape points around the obstacles. These escape points are virtual attractive points in the potential field that the robot can follow to overcome the obstacles more easily. The selection of which escape point to follow is done in real time by solving a soft-constrained problem optimizing the reaching of the most probable goal, estimated from the user’s action. Our proposal has been extensively compared with two state-of-the-art approaches in a static cluttered environment and a dynamic setup with randomly moving objects. Experimental results showed the efficacy of our method in terms of quantitative and qualitative metrics. For example, it significantly decreases the time to complete the tasks and the user’s intervention, and it helps reduce the failure rate. Moreover, we received positive feedback from the users that tested our proposal. Finally, the proposed framework is compatible with both mobile and manipulator robots.

This paper addresses the problem of mapping large surfaces with a moving sensor. In particular, it proposes image registration and mapping algorithms that enable to use in continuous motion sensors that need multiple shots to perform a measurement. These methods exploit the knowledge of the shape of the part to inspect in a both efficient and accurate fashion, thus allowing to obtain a measurement quality comparable to that of static measurements, while guaranteeing fast sensor motion and thus short scanning times. This work describes the application of these methods for the mapping of carbon fibre parts with an inspection robot equipped with a sensor estimating 3D carbon fibre orientation from multiple 2D images captured with different illumination. Experiments on carbon fibre preforms of complex 3D shape demonstrates that this system accurately reconstructs in real-time the 3D fibre orientations of the outer layer of carbon fibre parts. Accuracy assessments report small errors within the tolerances allowed by the automotive industry on flat and generic 3D surfaces. The inspection robot system presented in this paper has been demonstrated both as an in-line quality inspection robot for production of carbon fibre preforms and as a measurement device for improving the draping process in the prototyping of carbon fibre parts.

In this paper, we present a control system for the continuous teleoperation of a robotic manipulator via brain-machine interface (BMI). The proposed solution is based on shared control approach that allows the user to only focus on the operational tasks, while the low-level control details are automatically handled by the robotic intelligence. The user drives the manipulator through the imagination of limb movements (both hands vs. both feet) and a parameterized mapping function is implemented to convert the continuous BMI outputs into robot velocity commands which are sent to the shared control framework. The latter consists in: (i) a target predictor module, to infer the most probable target objects from the sequence of BMI commands; (ii) a control module based on an improved version of artificial potential fields (APF) to assist the user in reaching the target while avoiding collisions with obstacles in the environment. The system has been tested with a sample subject in a tabletop reach-to-grasp experiment with multiple target objects and obstacles achieving a success rate of 80%. The proposed system could be used in the future to help people with severe motor disabilities in performing daily life operations, such as drinking, feeding or manipulating objects.

Robots learning a new manipulation task from a small amount of demonstrations are increasingly demanded in different workspaces. A classifier model assessing the quality of actions can predict the successful completion of a task, which can be used by intelligent agents for action-selection. This paper presents a novel classifier that learns to classify task completion only from a few demonstrations. We carry out a comprehensive comparison of different neural classifiers, e.g. fully connected-based, fully convolutional-based, sequence2sequence-based, and domain adaptation-based classification. We also present a new dataset including five robot manipulation tasks, which is publicly available. We compared the performances of our novel classifier and the existing models using our dataset and the MIME dataset. The results suggest domain adaptation and timing-based features improve success prediction. Our novel model, i.e. fully convolutional neural network with domain adaptation and timing features, achieves an average classification accuracy of 97.3% and 95.5% across tasks in both datasets whereas state-of-the-art classifiers without domain adaptation and timing-features only achieve 82.4% and 90.3%, respectively.