Empirical Control System Development for Intelligent Mobile Robot Based on the Elements of the Reinforcement Machine Learning and Axiomatic Design Theory

Abstract

This paper presents the authors’ efforts to conceptual design of control system that can learn from its own experience. The ability of adaptive behaviour regarding the given task in real, unpredictable conditions is one of the main demands for every intelligent robotic system. To solve this problem, the authors suggest a learning approach that combines empirical control strategy, reinforcement learning and axiomatic design theory. The proposed concept uses best features of mentioned theoretical approaches to produce optimal action in the current state of the mobile robot. In this paper empirical control theory imparts the basis of conceptual solution for the navigation problem of mobile robot. Reinforcement learning enables the mechanisms that memorize and update environment responses, and combining with the empirical control theory determines best possible action according to the present circumstances. Axiomatic design theory accurately defines the problem and possible solution for the given task in terms of the elements defined by two previously mentioned approaches. Part of the proposed algorithm was implemented on the LEGO Mindstorms NXT mobile robot for the navigation task in an unknown manufacturing environment. Experimental results have shown good perspective for development of efficient and adaptable control system, which could lead to autonomous mobile robot behaviour.