Area Partitioning Method with Learning of Dirty Areas and Obstacles in Environments for Cooperative Sweeping Robots

In this paper, we introduce an extended performance-based partitioning method for the cooperative cleaning domain in the environment with obstacles. Due to ongoing advances in technology, robotic applications have been crucial for large and complicated areas that require cooperation and coordination in task operations by multiple robots. Therefore, our research has focused on methods for cooperation/coordination of multiple agents, which are the control programs of robots, using examples of cleaning tasks by multiple robots. Our proposed method partitions target area in a bottom-up manner, according to the characteristics of environments by identifying where are easy to be dirty, so that agents can clean their responsible areas effectively and evenly. Specifically, it also has included the learning to identify the shapes and the locations of obstacles in the environments via the steps of cleaning tasks because the shapes of obstacles affect the work performance. Our experiments showed that it could partition their responsible areas autonomously and effectively by taking into consideration the environmental characteristics. We also indicated that it could achieve efficient task operations in a more balanced manner by comparing these results with those by the conventional methods which assumed that the area is divided into equal-size sub areas and/or the environmental characteristics are given in advance.