Target differentiation using sonar data for robot applications; neural network approach

In this paper processing of sonar signals using data based approaches such as neural networks are used to differentiation of commonly met features in indoor robot environments is investigated. Amplitude and time-of-flight (TOF) characteristics of five various targets at some distances and angles are employed. Three types of neural networks are studied in different configurations. Also a useful configuration of modular neural network is developed to differentiate the objects. Performed comparisons between these approaches indicate high performance of using these types of data and methods for solving the problem of target differentiation for mobile robot applications.