Range Data Approximation for Mobile Robot by Using CAN2

In this article, we apply the competitive associative net called CAN2 to the processing of the range data of indoor environment acquired by a mobile robot, where the CAN2 is a neural net or a learning machine which performs piecewise linear approximation. After introducing several methods for dealing with range data by the CAN2, we show the following results; (1) an original range image involving lack of data or jump edges can be learned to be recalled as a natural range image by means of modifying the learning and recalling procedure of the CAN2, (2) high data compression ratio can be achieved the CAN2 although the quality of the range image is not reduced so much.