An mobile autonomous robot must have a reliable system of navigation that avoids large objects in its path such as recycling bins, etc., but can recognize ``important'' objects such as doorways. We have developed a system that can recognize such objects as doorways and recycling bins and that can predict the distance to these objects. The system is implemented by using artificial neural networks to classify objects and predict distances, supplemented by image segmentation of the camera image from the mobile robot into color regions.