Estimation of position of a mobile robot using a neural net

An active beacon localization system is developed for estimating the position of a mobile robot. Six beacons transmit a pattern of ultrasonic pulses. A receiver on the robot logs the arrival times of the pulses. This information is used to train a new single stage neural network (the cerebellar model with interpolation, or CEINT) to track the robot position. This network converges rapidly and has modest storage requirements. Its performance over the test set shows that a more extensive training data set would yield improved performance. This underlines the main disadvantage of using neural nets, i.e., their performance depends on the properties of the training set.<
>