Preliminary results of a novel approach for glucose regulation using an Actor-Critic learning based controller

In this paper, a novel approach for glucose regulation in individuals with type 1 diabetes mellitus is considered using the principles of adaptive reinforcement learning techniques. More specifically a controller based on Actor-Critic (AC) learning algorithm is used for the estimation of insulin infusion rate in adult patients using continuous glucose monitoring devices and insulin infusion pumps. Generally, AC algorithms are able to solve control problems of nonlinear, dynamic systems. Aim of the paper is to investigate the applicability of the AC learning method to the problem of glucose regulation and the feasibility to be used for an artificial pancreas. The implemented controller has been tested in an in silico environment. Preliminary results have shown that although the algorithm prevents hypoglycemic events, further research is needed in order to reduce the percentage of glucose concentration values over the acceptable normoglycaemia bound. Furthermore, open issues with the proposed algorithm have been identified, while the next research steps have been defined.