查看文章

In the mobile robotics field, localization and map building are very important tasks that a mobile robot must perform for a safe navigation. The present paper presents a multi-layer perceptron (MLP) neural classifier of obstacles with coded outputs. This means that the number of the network outputs is lower than the number of the recognizable patterns. In this way, the neural network is less cumbersome; this will make easier the network parametrizing and resizing to be applied to other type of environment. The developed classifier ensures the discrimination of all possible patterns using a little number of elements in the training set. Subsequently the problem of training slowness is avoided.