Thomas Agostinelli

Driver distraction remains a major road-safety challenge for Advanced Driver Assistance Systems and Driver Monitoring Systems, because attentional degradation is multidimensional and cannot be reliably captured by a single sensing modality. This study proposes a non-intrusive multimodal temporal pipeline for distraction detection, combining camera-derived driver-state descriptors with simulator-derived vehicle-control and road-context variables. Data were collected from 22 licensed drivers in a static driving simulator under two visual-distraction scenarios, yielding approximately 485,000 frame-level samples. Gaze, head pose, eye openness, arousal, engagement, vehicle dynamics, steering, pedal use, lane-keeping, and road-context features were synchronized, causally normalized, and segmented into 4 s temporal windows. Classification was performed using a dual-branch neural architecture with convolutional temporal encoding, self-attention in the core branch, and gated late fusion to condition the contribution of auxiliary contextual variables. Generalization to drivers not included in model training was assessed through Leave-One-Subject-Out cross-validation. The proposed model achieved an accuracy of 77.21\%, a macro-averaged F1 score of 0.7277, and a ROC-AUC of 0.8619. These results should be interpreted as conservative cross-subject estimates obtained under a strict participant-independent protocol. They suggest that causal temporal fusion of non-intrusive camera and vehicle-control signals can support window-based detection of driver distraction, while preserving a more realistic evaluation setting than random frame-level partitions.

Virtual Reality (VR) is increasingly adopted in cultural heritage to enable interactive exploration of artefacts that cannot be physically handled in museum settings. Despite growing interest in advanced interaction technologies such as haptic gloves and VR treadmills, empirical evidence on their combined effects on key user-experience metrics remains limited. This study investigates the individual and combined effects of haptic gloves and treadmill-based locomotion in immersive VR, addressing presence, usability, cognitive workload, and cybersickness. A within-subjects user study with 162 participants was conducted using a Unity-based VR application recreating a virtual jewelry museum. Four experimental conditions combined two interaction modalities (controllers vs. haptic gloves) with two locomotion techniques (controller vs. treadmill), all using an immersive HMD (HTC Vive Pro 2). Outcomes were measured via the Igroup Presence Questionnaire (IPQ), System Usability Scale (SUS), Instantaneous Self-Assessment (ISA), and Fast Motion Sickness (FMS) scales, and analyzed through linear mixed-effects models with the controller-only configuration as baseline. Results show that haptic gloves significantly decreased usability (β = −3.59, p < .05) and increased cognitive workload (β = 0.23, p < .05). Treadmill locomotion significantly reduced cognitive workload (β = −0.36, p < .05) but did not affect presence, usability, or cybersickness independently. A significant interaction emerged only for presence (β = −0.17, p < .05), indicating that combining gloves and treadmill yielded lower perceived presence than other configurations. Notably, cybersickness was not significantly affected by either the main effect of Interaction or Locomotion, nor by their combination. These findings suggest that advanced interaction technologies do not necessarily enhance immersion in VR museum experiences, and that cultural heritage applications should prioritize usable and cognitively efficient solutions, carefully weighing the benefits of haptic devices against their impact on usability and user comfort.

This study investigates the impact of Augmented Reality and Virtual Reality technologies on User Immersion and Learning Performance in museum settings. A quasi-experimental study, conducted in the context of an archaeological virtual museum, compares the effects of AR and VR interfaces on User Immersion and Learning Performance. Participants (n = 176) interacted with digital replicas of archaeological artefacts via AR and VR interfaces, followed by assessments using the Augmented Reality Immersion (ARI) questionnaire and knowledge-based multiple-choice tests. The use of the ARI questionnaire made possible a comprehensive investigation of the effect of User Immersion (and not only some of its aspects, e.g., Presence or Flow) on Learning Performance in the context of Cultural Heritage. The results revealed that AR users achieved higher Learning Performance, while VR users experienced a greater immersive experience. Significant correlations were found between higher User Immersion levels and lower Learning Performance. These findings highlight the importance for Museum curators to find a balance between exploiting technology to entertain and maintaining the educational integrity of the exhibits.

Appearance-based gaze estimation frequently relies on deep Convolutional Neural Networks (CNNs). These models are accurate, but computationally expensive and act as ‘‘black boxes’’, offering little interpretability. Geometric methods based on facial landmarks are lightweight alternatives. However, their performance limits and generalization capabilities remain underexplored in modern benchmarks. In this study, we conduct a comprehensive evaluation of the landmark-based gaze estimation. We introduce a standardized pipeline to extract and normalize landmarks from three large-scale datasets (Gaze360, ETH-XGaze, and GazeGene) and train lightweight regression models, specifically extreme gradient boosted trees and two neural architectures: a holistic multilayer perceptron (MLP) and a Siamese MLP designed to capture binocular geometry. We find that landmark-based models exhibit lower performance in the within-domain evaluation, likely due to noise introduced into the datasets by the landmark detector. Never-theless, in the cross-domain evaluation, the proposed MLP architectures show generalization capabilities comparable to those of the ResNet18 baselines. These findings suggest that sparse geometric features encode sufficient information for robust gaze estimation, paving the way for efficient, interpretable, and privacy-friendly edge applications. The source code and generated landmark-based datasets are available at: https://github.com/daniele-agostinelli/LandmarkGaze.git.

Generative Artificial Intelligence (GenAI) is widely adopted in design practice, yet its value in the testing phase remains underexplored. We present a five-phase methodology that conditions a multimodal large language model (MLLM) to role-play target users during videogame playtests. The phases are: (1) data acquisition, (2) user definition, (3) personas construction, (4) zero-shot role-play prompt design, and (5) iterative screenshot-grounded playtesting. We validate the approach with a point-and-click game in order to collect UI usability, UX impressions and game engagement feedback, comparing two data-grounded personas (“Instinctive” and “Methodical”) against six real playtesters. Using Jaccard similarity, the Instinctive persona achieved similarity to real users (0.42-0.76), comparable to human-to-human variability (0.48–0.71). Methodical persona was lower (0.22–0.42). Personas covered 80% of planned actions, detected 63.63% of users’ pain points, and reproduced 46.43% of “in-the-moment” feedback, but showed some limitations, like hallucination and loop behavior. Results indicate that data-grounded GenAI personas can complement, but not replace, human playtests, providing low-cost insights and early UX risk detection for resource-constrained teams.

This study contributes to understanding semi-automated ergonomic risk assessments in industrial manufacturing environments, proposing a practical tool for enhancing worker safety and operational efficiency. In the Industry 5.0 era, the human-centric approach in manufacturing is crucial, especially considering the aging workforce and the dynamic nature of the entire modern industrial sector, today integrating digital technology, automation, and sustainable practices to enhance productivity and environmental responsibility. This approach aims to adapt work conditions to individual capabilities, addressing the high incidence of work-related musculoskeletal disorders (MSDs). The traditional, subjective methods of ergonomic assessment are inadequate for dynamic settings, highlighting the need for affordable, automatic tools for continuous monitoring of workers’ postures to evaluate ergonomic risks effectively during tasks. To enable this perspective, 2D RGB Motion Capture (MoCap) systems based on computer vision currently seem the technologies of choice, given their low intrusiveness, cost, and implementation effort. However, the reliability and applicability of these systems in the dynamic and varied manufacturing environment remain uncertain. This research benchmarks various literature proposed MoCap tools and examines the viability of MoCap systems for ergonomic risk assessments in Industry 5.0 by exploiting one of the benchmarked semi-automated, low-cost and non-intrusive 2D RGB MoCap system, capable of continuously monitoring and analysing workers’ postures. By conducting experiments across varied manufacturing environments, this research evaluates the system’s effectiveness in assessing ergonomic risks and its adaptability to different production lines. Results reveal that the accuracy of risk assessments varies by specific environmental conditions and workstation setups. Although these systems are not yet optimized for expert-level risk certification, they offer significant potential for enhancing workplace safety and efficiency by providing continuous posture monitoring. Future improvements could explore advanced computational techniques like machine learning to refine ergonomic assessments further.

The digital games industry has grown exponentially due to the diversification of games and the increasing multiplicity of the user target base. The market explosion and the great variety make digital game cataloguing and classification challenging issues whose effectiveness can advance scientific research and address design, development and distribution. Firstly, the present study reviews previous cataloguing for video and serious games through systematic literature review and, joining together the findings from the literature review, develops a unified cataloguing model based on five definitions. This model can aid designers in tailoring their applications and contribute to disseminating game design knowledge in academic research. Then, a matrix that correlates design principles and the cataloguing model’s metadata is applied to the cataloguing model, obtaining a unified classification system. Together, they offer a comprehensive framework for understanding the multifaceted landscape of digital games, addressing the limitations of existing domain-specific approaches and providing a versatile tool for game designers. Research validation exploits a two-stage cluster analysis using agglomerative hierarchical and k-means clustering on the data extracted from a sample of digital games. The results show the framework's effectiveness in categorizing digital games without a clear-cut distinction between video and serious games. The system's application in real-world scenarios suggests its potential to guide game development. Future work will refine the proposal based on feedback from digital game designers, expanding the research scope to include a broader range of games.