MPC-guided Imitation Learning of Bayesian Neural Network Policies for the Artificial Pancreas

Although Model Predictive Control (MPC) is one of the main algorithms that has been proposed for insulin control in the context of artificial pancreas (AP), it typically requires complex online optimization, which is infeasible for resource-constrained medical devices. MPC also usually relies on state estimation, an error-prone process. In this paper, we introduce a novel approach to insulin control for the AP that uses Imitation Learning to synthesize neural-network policies from MPC-computed demonstrations. Such policies are computationally efficient and, by instrumenting MPC at training time with full state information, they can directly map measurements into optimal therapy decisions, thus bypassing state estimation. We apply Bayesian inference via Monte Carlo Dropout to learn policies, which allows us to quantify prediction uncertainty and thereby derive safer therapy decisions. We show that our control policies trained under specific patient models readily generalize (in terms of model parameters and disturbance distributions) to patient cohorts, consistently outperforming traditional MPC with state estimation.