Position Estimation of Mobile Robots Using Unsupervised Learning Algorithms

Abstract

Estimating the position of a mobile robot in an environment is a crucial issue. It allows the robot to obtain more precisely the knowledge of its current state and to make the problem of generating command sequences for achieving a certain goal an easier task. The robot learns the environment using an unsupervised learning method and generates a percept – action- percept graph, based on the readings of an ultrasound sensor. The graph is then used in the process of position estimation by matching the current sensory reading category with an existing node category. Our approach allows the robot to generate a set of controls to reach a desired destination. For the learning of the environment, two unsupervised algorithms FuzzyART neural network and GNG network were used. The approach was tested for its ability to recognize previously learnt positions. Both algorithms that were used were compared for their precision.