In this thesis, an inference method for understanding the intention of an obstacle for collision avoidance using the grid-type map is proposed. In order to represent the environment using ultrasonic sensors, the grid-type map is first constructed and fuzzy logic is used to treat the uncertainty and characteristics of ultrasonic sensors. Then, CLA(Centroid of Largest Area) points of the grid-type map are suggested to detect and avoid the obstacle; The CLA point is the safest point in each row of the grid-type map. Therefore, if there is no obstacle in front, CLA points make a straight line in the direction of which the mobile robot is heading.
Because an obstacle with the intention for collision avoidance tends to keep away from the path of the mobile robot, CLA points tend to move toward the path of the mobile robot. On the basis of this observation, the line regression is used in order to understand the tendancy of CLA points. The slope of the regressed line is a criterion to decide the intention of an obstacle for collision avoidance. To verify the proposed method, some experiments are performed.