Behavior Learning of Autonomous Robots by Modified Learning Vector Quantization

Abstract

This paper presents a method for searching for the optimalpaths for autonomously moving agents in mazes by modifiedLearning Vector Quantization (LVQ) in a reinforcementlearning framework. LVQ algorithm is faster than Q-learningalgorithms because LVQ concentrates on the best behavior inavailable behaviors while Q-learning algorithms calculatevalues of all available behaviors and choose the bestbehavior among them. However, ordinary LVQ sometimesmis-learns in the reinforcement learning environment due toerroneous teacher signals. In this paper a new LVQalgorithm is proposed to overcome this problem, which findsthe optimal path more efficiently.