Effective Method for Autonomous Simultaneous Localization and Map Building in Unknown Indoor Environments

Abstract

Many published papers [17, 44] on the map building of Autonomous Mobile Robots (AMRs) do not consider the question of autonomous exploration at all. This is, of course, often just a choice of research focus; effort is expended on the mechanics of map construction from sensor data without worrying about how the sensing positions were selected. Or the map is provided by the operator [7, 11] for any other applications. In our view, the autonomous exploration skill is an extremely important capability for a truly AMR. For example: as it is desired to build a map of unknown environments without human intervention, AMRs should be equipped with a skill of autonomous exploration which includes the competence of path finding, obstacle avoidance and monitor progress towards reaching a goal location or target. Several possible strategies for exploration of unknown environment are described in the robotics literature. The following categorization is taken from Lee [24]: 1. Human Control – mobile motion is controlled by human operator. 2. Reactive Control – the mobile robot movement is relied on the perception system. 3. Approaching the unknown – the mobile robot move into the region that it knows least in the environment. 4. Optimal search strategies – the approach is focused on to search the shortest path for seeking the goal. In the first category, the robot is guided around the environment by a human operator. This requires human intervention in the map building process. Therefore, it is not suitable for an autonomous exploration mobile robot. For reactive exploration approach (2nd category), the sensory data (perception space) is used to calculate or determine the control actions (action space). The sensory data may be the distance information from infrared, sonar or laser range finder type sensors, visual information or processed information obtained after appropriate fusion of multiple sensor outputs. The control actions are usually a change in steering angle and setting a translation velocity of the robot that will avoid collisions with the obstacles on its way and reach the desired target. Predesigned or adaptive systems based on fuzzy logic [15, 26, 28, 31, 40-41, 45-46, 48-49], neuralnetworks [9, 33-35, 38, 51] or combination of them [27] are designed by this reactive navigation