Frontier-based exploration is the most common approach to exploration, a fundamental problem in robotics. In frontierbased exploration, robots explore by repeatedly detecting (and moving towards) frontiers, the segments which separate the known regions from those unknown. A frontier detection sub-process examines map and/or sensor readings to identify frontiers for exploration. However, most frontier detection algorithms process the entire map data. This can be a time-consuming process, which affects the exploration decisions. In this work, we present several novel frontier detection algorithms that do not process the entire map data, and explore them in depth. We begin by investigating algorithms that represent two approaches: Wavefront Frontier Detector (WFD), a graph-search-based algorithm which examines only known areas, and Fast Frontier Detector (FFD), which examines only new laser readings data. We analytically examine the complexity of both algorithms, and discuss their correctness. We then improve by combining elements of both, to create two additional algorithms, called WFD-INC and WFD-IP. We empirically evaluate all algorithms, and show that they are all faster than a state-of-the-art frontier detector implementation (by several orders of magnitude). We additionally contrast them with each other and demonstrate the FFD and WFD-IP are faster than the others by one additional order of magnitude.
Bibliographical noteFunding Information:
This research was supported in part by IMOD and ISF grant number 1511/12.