A robust tube-based MPC approach for safer insulin dosing

We present a robust model predictive control (MPC) approach for safer insulin dosing in artificial pancreas systems. The method applies estimation-informed constraint tightening to maintain safety under disturbances and unmodeled dynamics. Safety margins, derived from estimator confidence, are enforced as tube-based tightening on state and input constraints. A zone control objective targets a glucose band with asymmetric weights that emphasize avoidance of low glucose. Simulation studies show fewer low glucose events and lower violation risk than a nominal MPC baseline, with comparable dosing effort. The contribution of this work is a robust zone MPC scheme that maps estimator uncertainty into safety margins for insulin dosing.