Hybrid control for robot navigation - A hierarchical Q-learning algorithm

Abstract

A control approach, hybrid control architecture, combines reactive and deliberate control using a hierarchical Q-learning algorithm. Under this control approach, the grid-topological maps are constructed and maintained online providing a model of the environment, and then, hybrid control architecture is proposed based on the grid-topological maps. Another novel q-learning algorithm is discussed based on the hybrid Markov decision process. This process runs as an integrated learning and control algorithm for the proposed hybrid control architecture. In addition, a grid-topological map-building method has been presented with online updating techniques.