Adaptive control in anaesthesia

Automation of anaesthesia is a complex task but important with respect to patient health, improved quality of narcosis and cost reduction. Furthermore it will enhance our understanding of the complex mechanism underlying anaesthesia. Classical model based control concepts have been evaluated in the past. Those approaches were limited to univariate process control. As the objective of our studies we want to establish the feasibility of different real-valued reinforcement learning approaches for the task of multivariate adaptive control in anaesthesia. As a first step we present a series of experiments with a naive application of reinforcement learning. The appropriateness is demonstrated in the univariate case. Results are compared to a model based analytical controller