Embedded multimodal physiological platforms with edge AI: a cross-domain journey from wearables to driver monitoring

The growing demand for continuous physiological monitoring in both healthcare and everyday contexts has driven the development of increasingly integrated and autonomous sensing systems. Recent advances in sensing technology and ultra-low-power electronics have made it feasible to acquire multiple biosignals unobtrusively and continuously, bringing long-term monitoring closer to everyday use. At the same time, the rise of artificial intelligence (AI) has enabled the transformation of raw sensor data into meaningful health and behavioral insights. However, executing such algorithms in the cloud introduces latency and privacy concerns, limiting responsiveness and autonomy. These challenges have accelerated the shift toward edge AI, where computation is performed locally on the device to enable real-time, context-aware, and energy-efficient operation. Achieving this goal requires a tight hardware–software co-design to balance accuracy, latency, and energy efficiency within compact, low-power systems. Interestingly, similar technological trends that have advanced wearable systems are now transforming the automotive domain. Modern driver monitoring systems (DMS), increasingly mandated by safety regulations such as the EU Driver Drowsiness and Attention Warning (DDAW) requirement, aim to detect fatigue, distraction, and cognitive load in real time. Current DMS implementations predominantly rely on camera-based behavioral cues, which are often sensitive to lighting conditions and occlusions, providing only indirect estimates of driver state. Integrating physiological monitoring offers a more direct and robust means to assess alertness and stress, but embedding such sensing within the vehicle cockpit demands unobtrusive design, resilience to motion artifacts, and efficient on-board computation. These shared challenges unify wearable and automotive research under a common goal: enabling multimodal, energy-efficient physiological inference at the edge. This thesis advances this vision through two complementary research directions: the development of a wearable edge-intelligent platform for multimodal biosignal acquisition, and the design of embedded driver monitoring systems based on biosignals and capable of assessing physiological states.