Hidden Markov Model for Control Strategy Learning

Abstract

This report presents a method for learning a control strategy using the hidden Markov model (HMM), i.e., developing a feedback controller based on HMMs. The HMM is a parametric model for non-stationary pattern recognition and is feasible to characterize a doubly stochastic process involving observable actions and a hidden decision pattern. The control strategy is encoded by HMMs through a training process. The trained models are then employed to control the system. The proposed method has been investigated by simulations of a linear system and an inverted pendulum system. The HMM-based controller provides a novel way to learn control strategy and to model the humans decision making process.