Active SLAM and Loop Prediction with the Segmented Map Using Simplified Models

We previously introduced the SegSLAM algorithm, an approach to the simultaneous localization and mapping (SLAM) problem that divides the environment up into segments, or submaps, using heuristic methods. We investigate a realtime method for Active SLAM with SegSLAM, in which actions are selected in order to reduce uncertainty in both the local metric submap and the global topological map. Recent work in the area of Active SLAM has been built on the theoretical basis of information entropy. Due to the complexity of the SegSLAM belief state, as encoded in the SegMap representation, it is not feasible to estimate the expected entropy of the full belief state. Instead, we use a simplified model to heuristically select entropy-reducing actions without explicitly evaluating the full belief state. We discuss the relation of this heuristic method to the full entropy estimation method, and present results from applying our planning method in real-time onboard a mobile robot.