Efficient Information Sharing and Coordination in Cooperative Multi-Robot Systems Rui Rocha Institute of Systems and Robotics University of Coimbra, Pole 2 3030-290 Coimbra, PORTUGAL E-mail: rprocha@isr.uc.pt Abstract— Multi-robot systems involve the distribution of robotic resources and information. Being an opportunity, they require cooperation among robots so that potential advantages of distribution become effective. Cooperation requires in turn efficiently sharing information and proper coordination. This article extends previous work, regarding efficient information sharing in the context of volumetric mapping [1], with a coordinated exploration method based on mutual information minimization. Experimental data obtained with multi-robot systems varying in size demonstrate the performance gain due to the proposed coordination method. Index Terms— Multi-robot systems, cooperation, coordination, information utility, 3-D mapping, exploration. I. I NTRODUCTION Multi-robot systems (MRS) are sets of autonomous mobile robots that are assumed to cooperate in order to carry out collective missions [2], [3]. MRS may either substitute humans in risky scenarios [4]–[7] due to the expendability of individual robots, or relieve people from collective tasks that are monoto- nous and repetitive. Moreover, they allow to automate missions that are inherently distributed in time, space or functionality. The distribution of robotic resources and information en- dows MRS with interesting features, such as space and time distribution, managing complexity through distribution, distri- bution of risk and increased robustness [8]. But these potential advantages require robots’ cooperation in order to become effective [9]. Cooperation requires in turn efficient information sharing and proper coordination. Previous work was conducted with the aim of restricting communication in MRS to useful information, in the context of building volumetric maps [1]. This framework is extended herein with coordinated multi- robot exploration, so as to improve collective performance. 1) Sharing information within multi-robot systems: Most of the work about MRS has been devoted to the definition of different distributed architectures [4], [6], [10] that rule the interaction between the behaviors of individual robots. Although communication is a central issue of MRS, because it determines the possible modes of interaction among robots, it has been often neglected in these architectures. Furthermore, most of the work about communication in MRS [10]–[13] has addressed the communication structure, neglecting another important dimension: the communication content. A distributed group architecture for 3-D mapping was proposed in [1], which endows robots with an altruistic infor- mation sharing behavior, wherein communication efficiency is ensured by restricting communication contents to useful infor- mation. This framework is extended herein with coordination. 2) Robotic mapping: Robotic mapping addresses the prob- lem of acquiring spatial models of physical environments with mobile robots equipped with range sensors. It is a relevant application domain whether robots are used to build detailed maps of environments, especially hazardous environments for human beings [5], [7], or they require a map to safely navigate within the environment and perform other useful tasks. As sensors have always limited range, are subject to oc- clusions and yield noisy data, mobile robots have to navigate through the environment and build the map iteratively. Key challenges include the sensor modeling problem, the repre- sentation problem, the registration problem and the exploration problem [14]. This article focuses on coordinated exploration. 3) Exploration and active sensing: When a robot or a team of robots explore an unknown environment to build a map, the main goal is to acquire as much new information as possible with every sensing cycle, so as to minimize the mission time. Yamauchi et al. proposed frontier-based exploration [15] whereby a robot is driven to the closest frontier cell in its neighborhood, located between open space and unexplored regions. Burgard et al. used this concept to address coordi- nation in multi-robot exploration, by considering a balance between travel cost and utility of unexplored regions, so that robots explore non-overlapping regions [16]. Bourgault et al. [17] addressed the single robot exploration problem as a balance of alternative motion actions, from the point of view of information gain (in terms of entropy), localization quality (using SLAM) and navigation cost. In [18], robots that can communicate with each other are arranged in exploration clusters and the robots within each cluster share a common map and coordinate their exploration actions as in [16]. This article proposes a multi-robot exploration method closely related with [16], but with important improvements: it uses a distributed architecture model with efficient information sharing [1], wherein entropy is used to define a formal information-theoretic background to reason about the mapping and exploration process; and the utility of an exploration viewpoint is formally defined using entropy-related concepts.