A Randomized Strategy for Cooperative Robot Exploration Antonio Franchi Luigi Freda Giuseppe Oriolo Marilena Vendittelli Dipartimento di Informatica e Sistemistica Universit` a di Roma “La Sapienza” Via Eudossiana 18, 00184 Roma, Italy {franchi,freda,oriolo,venditt}@dis.uniroma1.it Abstract— We present a cooperative exploration strategy for mobile robots. The method is based on the randomized incremental generation of a collection of data structures called Sensor-based Random Trees, each representing a roadmap of an explored area with an associated safe region. Decentralized cooperation and coordination mechanisms are introduced so as to improve the exploration efficiency and to avoid conflicts. Simulations in various environments are presented to show the performance of the proposed technique. I. I NTRODUCTION Exploration is the basic task by which a mobile robot covers an unknown area, typically learning a model of the environment at the same time. Possible applications include automated surveillance, search-and-rescue operations in hostile areas, map building and planetary missions. The use of a multi-robot system brings in general many advantages [1], [2]. In exploration, it aims at significantly reducing the time required to complete the task. If a map is to be acquired, the redundant information provided by multiple robots can be also used to increase the final map accuracy and the quality of the localization [3]. In order to achieve these objectives, some sort of task decomposition and allocation are required. In practice, strategies to conveniently distribute robots over the environment should be accurately devised in order to reduce the occurrence of spatial conflicts [4] and actually reap the benefits of a multi-robot architecture. Clearly, communication plays a crucial role in achieving a cooperative behavior with improved performance [5]. In most exploration strategies, the boundary between known and unknown territory (the frontier) is approached in order to maximize the information gain. For the multi- robot case, a pioneering work in this direction is [6]: the robots merge the acquired information in a global gridmap of the environment, from which the frontier is extracted and used to plan the individual robot motions. While this basic scheme lacks an arbitration mechanism preventing robots from approaching the same frontier region, in [7] it is proposed to negotiate robot targets by optimizing a utility function which takes into account the information gain of a particular region, the cost of reaching it and the number of robots currently heading there. In [8], the utility of a particular frontier region from the viewpoint of relative robot localization (and hence, of the accuracy of map merging) is also considered. In the incremental deployment algorithm of [9], robots approach the frontier while retaining visual contact with each other. An interesting multi-robot architecture in which robots are guided through the exploration by a market economy is presented in [10], whereas [11] proposes a centralized approach which uses a frontier-based search and a bidding protocol assign frontier targets to the robots. This paper presents a randomized strategy for cooperative exploration which is the outgrowth of the SRT method, designed for a single robot and presented in [12], [13]. The basic tool therein is the Sensor-based Random Tree (SRT), a compact data structure representing a roadmap of the explored area, which can be seen as a sensor-based version of the RRT concept [14]. In particular, each node of an SRT contains a configuration assumed by the robot and the Local Safe Region (LSR) perceived from that location, while an arc between two nodes represents a collision-free path between the two configurations. The SRT is incrementally built by using a randomized local planner which privileges the frontier of the LSR, i.e., the directions that lead from the LSR to unexplored areas. This mechanism automatically realizes a trade-off between information gain and navigation cost when choosing the next robot configuration. Our cooperative exploration strategy is essentially a par- allelization of the single-robot SRT method, with the addi- tion of three functionalities: (i) cooperation, for increasing efficiency (ii) coordination, to avoid conflicts (iii) commu- nication, as a fundamental tool to cooperate and coordinate. Each robot of the team builds an SRT, taking into account the presence of other robots through an appropriate redefinition of the concept of local frontier, and planning its motion toward areas which appear to be unexplored by itself as well as the rest of the team. In addition to this local cooperation mechanism, there is a simple coordination algorithm that guarantees safe collective motion. Once a robot has com- pleted its SRT, it makes itself available for supporting other robots in their expansion; this introduces a form of global cooperation. A key feature of the proposed strategy is that it is completely decentralized and can be implemented with a limited communication range. The paper is organized as follows. After stating the working assumptions, we outline in Sect. III the basic steps of the exploration algorithm running on each robot. In particular, the construction of an SRT with the associated 2007 IEEE International Conference on Robotics and Automation Roma, Italy, 10-14 April 2007 WeC1.3 1-4244-0602-1/07/$20.00 ©2007 IEEE. 768