A neural network approach to robot localization using ultrasonic sensors

A regression-based approach is suggested for solving the task of robot localization using ultrasonic sensing. The regression is performed by using an artificial neural network approach, with the advantage that no explicit regression modeling is required. The use of entropy net methodology to implement neural regression is suggested. The advantage of the entropy net methodology is that it yields the structure of the network through a data-driven process that first obtains a tree structure for the problem. In addition to providing the network structure, the regression tree also provides an insight into the various relationships present in the problem. Details of the localization tasks and experimental results are provided