Leader: Paolo Barsocchi (CNR); Other collaborator(s):
This task will investigate a new generation of smart spaces in multi-resident scenarios, enriched with novel devices and AI tools to automatically detect early signs of physical and cognitive decline through physiological and behavioral data. Among the main sources of information, indoor localization and vital signals monitoring represent the most challenging research fields. Novel solutions need to be investigated to overcome the lack of a de facto standard and the trade-off between performance and complexity/costs of indoor localization systems. Furthermore, the development of novel unobtrusive sensors and the demand for personalized interaction experience still need to be addressed for the adoption of vital signs monitoring solutions.
During the reporting period, the investigation involved a comprehensive analysis of indoor localization and gait analysis systems. This include identifying the requisite hardware and software equipment essential for data collection and analysis purposes. In particular, the proposed activities explore technologies for estimating user location, including BLE and UWB, with a focus on RSS, AoA, and ToF. Stability is analyzed using sensor mats, with AI algorithms developed for early detection based on stability and mobility characteristics. The use of simulated, emulated, and real-world data is planned for thorough analysis. Additionally, a thorough review of the existing literature provided valuable insights into relevant methodologies and approaches. Furthermore, a detailed description of the preliminary research methodology undertaken was presented, offering a comprehensive overview of the initial stages of the study.
Please see the next reporting period.
We concentrated on utilizing Bluetooth Low Energy (BLE) signals for proximity detection, leveraging their widespread availability in commercial devices like smartphones and wearables. The goal was to refine methods for determining proximity between users and specific points of interest (POI) based on Received Signal Strength (RSS) values. This approach is not limited to a specific scenario but is designed to be applied in various contexts where proximity detection is needed, such as indoor navigation, smart buildings, and healthcare environments.
By analyzing BLE signal strengths and employing learning-based algorithms alongside threshold-based methods, our system can accurately determine when a user is near a particular location or object. The technique leverages crowdsourced data to improve calibration over time, making it adaptable to different physical settings and user devices. This general framework can be integrated into any application requiring proximity sensing, providing a flexible and scalable solution for environments where precise indoor positioning is not necessary, but knowing the closeness of users to certain areas is critical.
In this period, we are focusing on utilizing BLE signals to identify the angle of arrival of the transmitted signal. This method is particularly effective for the new generation of smart spaces, where the goal is to localize and identify a user within a given environment. To support this, we conducted an experimental campaign to collect BLE signals in an office environment. We are currently studying methodologies to infer the user's position within this indoor environment, aiming to improve localization accuracy and efficiency in such settings.
During the reporting period, the investigation focused on a comprehensive analysis of indoor localization and gait analysis systems. Specifically, the activities explored technologies for estimating a user's position based on the Angle of Arrival (AoA) of Bluetooth Low Energy (BLE) signals. In parallel, gait stability was analyzed using sensor mats, and AI algorithms were developed to enable early detection of potential issues by evaluating stability and mobility characteristics.
During the current reporting period, the work progressed towards the development of intelligent monitoring systems by extending the previous investigations on indoor localization and gait analysis. Building on the previously analyzed BLE-based AoA positioning and gait stability assessment through sensor mats, the activities focused on studying models for automatic anomaly detection. These models aim to identify irregular mobility patterns and stability deviations in real time, leveraging the data collected from the aforementioned systems.
During the reporting period, the research activities focused on advancing AI-driven methods for indoor positioning and navigation. Two main research directions were followed.
On the one hand, we developed and evaluated a novel framework based on Physics-Informed Neural Networks (PINNs) for indoor localisation using Bluetooth Low Energy (BLE) signals. The study demonstrated how integrating physics-compliant synthetic data during training can significantly reduce the need for large real-world datasets, while maintaining high accuracy and improving model generalization. On the other hand, we explored Large Language Models (LLMs) for intelligent indoor navigation. The proposed system leverages ChatGPT to interpret indoor map images and generate natural, context-aware navigation instructions. Experimental evaluation across real-world settings achieved high accuracy, confirming the potential of multimodal LLMs for assistive navigation.
The activities carried out in this period strengthened the project’s objectives by combining physics-based learning and generative AI for enhanced human-centered indoor sensing and navigation systems.
Scientific Publications