Fusing and modelling sensors using neural networks for low cost multi-sensor robotics systems

Abstract

In this paper we deal with the problem of modeling the sensory behavior of sensors in a low cost robotics system, modeling the uncertainty of taken measurements and fusing different measurements, in order to improve the quality of the information that a robotics system needs. It is achieved thanks to a sensory model based on multilayer perceptrons, trained off-line. This model not only improves the quality of the information, but also grants independence between hardware and software of the robotics system. Some experiments have been done, using the mobile robot UMI, in order to demonstrate the utility of this model. These experiments show how sensory model corrects the ultrasonic reflection effect and the sliding of wheels.