LAUSR

Dear editor, Multi-robot exploration of indoor environments is a fundamental problem in mobile robotics [1]. In the context of multi-robot exploration, the main challenge is in achieving coordination among the robots such that they can be better distributed over the environment for simultaneously exploring different regions and avoiding interference with other robots [2, 3]. To date, many approaches have been proposed to address this issue. However, most of these approaches are only focused on the applications that assume zero prior knowledge about the environments, and very limited studies consider the additional background knowledge or any assumptions about the environmental structure. In general, indoor environments constructed by humans often contain certain structures, which are presented in terms of typical semantic concepts like corridors, doorways, offices, seminar rooms. By taking advantage of such rich semantic information, robots can perform exploration tasks more efficiently. In this study, we present a novel frontier-based exploration strategy to coordinate multiple mobile robots for exploring indoor environments, which aim to minimize the time needed to explore the whole environment. Our coordination technique is motivated by the studies of [4,5], in which Ref. [4] gives a rigorous proof that by maximizing the expected utility and by minimizing the potential for overlap in information gain, targets can be appropriately assigned to the individual robots, and Ref. [5] shows that the supervised learning approaches could be used to estimate the background knowledge about environment structure and incorporate it into the target assignment procedure. Proposed multi-robot exploration strategy. To assign appropriate target frontiers for the individual robots, we present a decision-theoretic exploration approach for explicitly coordinating the mobile robots based on the iteratively evaluating a number of target frontiers according to the costs for moving the target frontiers and their utilities during the target frontier assignment procedure, which determines where the robots would move next. Furthermore, we take into account the semantic information of target frontiers and integrate this knowledge into the utility functions to help the robots obtain a higher reward for exploring non-corridor places, such as offices, seminar rooms, and study rooms. As a result, the robots have higher priority to thoroughly explore these rooms and will not have to return to the previously explored places, which leads to a shorter overall exploration trajectory and thus reduces the total exploration time. Figure 1 illustrates the schematic view of our proposed multi-robot exploration system, which