Multi-Step Optimal Tracking Control of Unknown Nonzero-Sum Games based on Least Squares and Linear Programming: An Application to a Fully-Automated, Dual-Hormone Artificial Pancreas

Abstract

We consider the problem of optimal tracking control of unknown discrete-time nonlinear nonzero-sum games. The related state-of-art literature is mostly focused on Policy Iteration algorithms and multiple neural network approximation, which may lead to practical implementation challenges and high computational burden. To overcome these problems, we propose a novel Q-function-based multi-step Value Iteration algorithm, which provides the potential to accelerate convergence speed and improve the quality of solutions, with an easy-to-realize initialization condition. A critic-only least squares implementation approach is then employed, which alleviates the computational complexity of commonly used multiple neural network-based methods. Afterwards, by introducing the coupled Bellman operator, a novel linear programming approach is derived, based on which Nash equilibria can be approximately computed by solving a set of tractable finite-dimensional optimization problems. We evaluate the tracking control capabilities of the proposed algorithms to the problem of fully-automated dual-hormone (i.e., insulin and glucagon) glucose control in Type 1 Diabetes Mellitus. The U.S. FDA-accepted DMMS.R simulator from the Epsilon Group is used to conduct extensive in-silico clinical studies on virtual patients under a variety of completely unannounced meal and exercise scenarios. Simulation results demonstrate the high reliability and exceptional performance of the proposed multi-step algorithmic framework to critical complex systems.