Multi-Modal Active Sensing for a Simple Mobile Agent

This work presents a methodology for the combination and use of many simple features of di ering modalities that are extracted from the environment. There are two main ideas presented here, rst that it is important to consider all modalities for extracting sensory features and second, that useful information can be extracted without resorting to computationally expensive methods. Many useful features can be extracted easily and cheaply and recognition robustness could be achieved through the combination of many such simple features. Examples of such feature parameters are based on color, texture, motion, range, sound, and magnetism. A design is given for a small, simple, and fast embodied agent employing the ideas contained herein.