Monitoring and Information Fusion for Search and Rescue Operations in Large-scale Disasters Fabrizio d’Agostino fdagosti@dis.uniroma1.it Alessandro Farinelli farinelli@dis.uniroma1.it Giorgio Grisetti grisetti@dis.uniroma1.it Luca Iocchi iocchi@dis.uniroma1.it Daniele Nardi nardi@dis.uniroma1.it Dipartimento di Informatica e Sistemistica Università di Roma "La Sapienza" Via Salaria 113, 00198 ROMA, Italy Abstract - The goal of the project, which is currently under development, is to design tools to monitor the situation after a large-scale disaster, with a particular focus on the task on situation assessment and high-level information fusion, as well as on the issues that arise in coordinating the agent actions based on the acquired information. The development environment is based on the RoboCup-Rescue simulator: a simulation environment used for the RoboCup-Rescue competition, allowing for the design of both agents operating in the scenario and simulators for modeling various aspects of the situation including the graphical interface to monitor the disaster site. Our project is focussed on three aspects: modeling in the simulator a scenario devised from the analysis of a real case study; an extension of the simulator enabling for the experimentation of various communication and information fusion schemes; a framework for developing agents that are capable of constructing a global view of the situation and of distributing specific information to other agents in order to drive their actions. Keywords: disaster simulation, multi-agent systems, information acquisition, situation assessment, information fusion, planning. 1 Introduction Search and rescue of victims in large-scale disasters are not only highly relevant social problems, but pose several challenges from a scientific standpoint. When earthquakes, eruptions or floods happen, a considerable organizational capability to aid the disaster victims as fast as possible is required. However, too often different secondary disasters, connected with the main one, occur, which avoid the correct execution of a rescue plan a priori decided. For example, as reported for the Kobe earthquake (1997) [12], after the earthquake, fires arose between the debris of the destroyed wooden houses, communication infrastructures and transportation systems were largely damaged, causing additional difficulties for the aids. The goal of the present project is to develop software tools to support the management of this kind of emergency, more specifically to design a support system for search and rescue operations in large-scale disasters, both for prevision and training as well as for operative actions. Even though there are significant results on the design of robots to support search and rescue [14], here we are specifically concerned with the design of a software tool to support both the situation assessment and the planning/control of operations; moreover, we are interested in the deployment of the techniques for achieving cooperation in a multi-agent system [18]. In this context, the RoboCup initiative, the organization which, starting from 1997, arranged the world championship for soccer player robots [9], proposed a new scientific challenge named RoboCup-Rescue [10, 11, 19], where the problems faced are those of bringing aids in a large disaster. The RoboCup objective is to create system based on AI and Robotics technologies, where heterogeneous agents (software, robots, human beings) interact in a cooperative manner. In particular, we are concerned with the possibility of deploying and extending the RoboCup-Rescue Simulator [17], a simulation system whose main feature is the ability to deal simultaneously with many events, thus allowing for the study of different rescue strategies with agents, autonomous or not, and to facilitate the development decision support systems working in real-time. There are several technical issues that arise in order to pursue the design of a tool like the RoboCup-Rescue simulator: event modeling and simulation, integration and visualization of data, resource management, planning and scheduling, execution monitoring. We are specifically interested in the problems arising in the interaction of a large number of heterogeneous agents [18], i.e. the members of a rescue team. Another relevant research aspect, strictly related to the previous one, is concerned with the coordination of the agents themselves, while they are trying to achieve a common goal [3, 6]. In particular,