Michele De Carolis

Dyson spheres are hypothetical megastructures proposed to partially or fully enclose a star and reprocess its radiation, producing mid-infrared excesses that may appear as rare anomalies in large stellar catalogues. No confirmed Dyson sphere is known, but the systematic prioritization of stars with unexplained photometric behaviour offers a practical way to support targeted astrophysical follow-up. In this work, we present a machine learning framework for the prioritization of potential Dyson sphere candidates, combining a two-phase design with normal and anomalous star models. Isolation Forest delivers the most accurate and stable overall ranking. The pipeline recovers known excess stars and highlights new candidates for follow-up.

Mobile robots operating in clinical and assistive environments require the ability to monitor human physiological states without direct contact. While remote photoplethysmography (rPPG) enables the estimation of vital signs from video, existing approaches focus on signal extraction and rarely address downstream clinical decision-making. In this work, we present an AI-driven multimodal pipeline that bridges contactless physiological perception and early warning risk assessment on a mobile robotic platform. The system is deployed on a TIAGo Pro robot equipped with an Intel RealSense D435 sensor, enabling synchronized RGB, near-infrared, and depth acquisition. A multimodal pipeline combines depth-stabilised region-of- interest selection, adaptive RGB-NIR fusion for cardiac estimation, and a depth- based respiratory channel, aiming to provide robust estimates of heart rate, respiratory rate, and oxygen saturation. Building on these measurements, we introduce an interpretable early warning risk assessment module inspired by clinical scoring systems, designed to operate under uncertain sensing conditions. The system is designed as a ROS 2 perception node and will be evaluated through a staged validation plan. This work advances risk-aware embodied perception for contactless monitoring, enabling mobile robots to support continuous and non-invasive patient assessment in real-world scenarios.

Runtime monitoring of ROS 2 systems remains an open problem: existing tools provide point-in-time snapshots without automated detection, and academic approaches operate offline or in simulation. We propose a framework combining ros2_tracing-based low- overhead data collection with a three-level self-training anomaly detection.

Ensuring safety in Physical AI systems operating in real-world environments is a critical challenge, particularly in hospital wards where vulnerable patients, clinical staff, medical devices, and assistive robots coexist. In this paper, we reinterpret Clinical Pathways as explicit runtime safety specifications for embodied medical AI. We propose a conceptual robotic architecture that integrates wearable sensors, smart medical devices, and assistive robotic components into a unified framework for real-time safety monitoring. At its core, a Runtime Safety Monitor (RSM) evaluates multimodal physiological and system-level signals against clinically defined constraints derived from the prescribed care process. Rather than relying solely on statistical anomaly detection, the proposed approach combines temporal prediction, uncertainty-aware reasoning, and constraint-based verification to identify safety violations. The RSM targets three classes of events: physiological deviations from prescribed care, hardware and communication failures, and potential data tampering or misuse. This work contributes to Safe Physical AI by operationalizing domain-specific clinical knowledge as enforceable safety constraints, bridging learning-based perception and runtime safety monitoring to assist nursing staff in real-world hospital wards.