“…Our controller is an MPC algorithm that addresses these challenges by handling the stochastic CGM and patient variations using a Kalman filter, by controlling the glucose level to a range rather than a specific glucose value, by providing steady state offset free glucose control, by being robust against patient-model mismatch by its definition of target trajectories and use of low-order models, by being individualized to each patient by use of already available parameters, and by including medical expert rules for safety purposes. The controller includes, combines, and extends key principles from model-based proportional-integral-derivative controllers, 23 previous MPC algorithms, [24][25][26][27][28] fuzzy logic controllers, 29 and novel theoretical MPC insights. 30,31 Intravenous glucose was administered on three CL study nights (Figure 2).…”