Safe Glucose Regulation in Artificial Pancreas Systems via Robust Control Barrier Functions

Abstract

Artificial pancreas systems (APSs) have transformed insulin delivery for individuals with Type 1 Diabetes Mellitus. However, these systems remain vulnerable to both physiological disturbances (meal) and cyber-physical threats such as Denial-of-Service (DoS) attacks that block insulin delivery actuation channel. This work presents a robust control framework using control barrier functions (CBFs) to enforce glucose safety constraints without violating hyper- or hypoglycemia, even under attack-induced uncertainties. We develop both standard and robust CBFs based on the physiological model of glucose-insulin interactions. We introduce a novel insulin-adaptive safety filter that anticipates delayed insulin action for improved closed-loop response. The proposed method guarantees for forward invariance and input-to-state safety. To counteract actuation-layer DoS attacks, we propose a fallback defense mechanism. Simulation results demonstrate the effectiveness of the proposed approach in maintaining glucose levels within safe bounds, ensuring resilience and reliability in APS operation.