Active Perception in SLAM Exploration: Problem Formulation and Methods Review

Active SLAM (A-SLAM) aims to enable an agent’s autonomous exploration and mapping task, provide robustness and accuracy in both localization and mapping and enhance environmental adaptivity in applications. Its working principle, as a holistic probabilistic-based problem in model formulation, can be divided into three decoupled working stages and accordingly solved, these stages are brief action identification, cost and utility computation, and action execution. The identification stage needs reasoning on what and where to look at in order to diminish uncertainty in an efficient fashion, by employing sensory methods and performing searching or optimization tasks with appropriate environmental representations, to feed the following stages with a discretized set of planning goals. This is the active perception problem by definition. The activeness of active SLAM is mostly decided at the perception stage with appropriate quantifying measures. This paper revisits the formulation of the A-SLAM problem especially by the stage of active perception, reviews its environmental representations (maps) and related modeling measures for information and uncertainty, underlines a sample-based aim of maximizing the information gain from sensors with respect to corresponding maps, and make comments on some novel approaches.