Integrating Linguistic Descriptions and Sensor Observations for the Navigation of Autonomous Robots

Abstract

One of the main problems for autonomous robot navigation in unknown indoor environments is the difficulty identifying the objects in the robot’s working area from raw sensor observations. Range sensors (i.e., laser and ultrasonic sensors), the most commonly used type of sensors in mobile robot applications, only provide information about the existence of objects in some given positions of the space. A number of approaches [2,17,20,24] have shown that this information is enough to successfully perform obstacle avoidance even in cluttered environments. However, there are many applications that also require to identify the type of objects that have been detected in order that the robot can take appropriate decisions and operate on the environment. The transition from range data to object identification is a very difficult problem, particularly when no a priori environment knowledge is available.