Reasoning about situations in the early post-disaster response environment
A prerequisite to efficient behavior by a team of agents or robots is the ability to accurately interpret the situation in which action must take place. Although individual agents, with local and uncertain information by sensors, will not be able to have enough information about the environment, to take correct decisions, collectively the team may be able to. This problem is known as Situation Assessment, and is usually solved relying on a centralized decision maker, or using massive communications among the agents. This research presents the steps to build an approach to cooperatively classify the situation at hand, specifically focused on improving the choice of team action. Specifically, agents (or robots) interpret locally the situation at hand using state-of-the-art high level reasoning based upon a Description Logics framework. Then, debate their local conclusions in a fully distributed manner, attaching only relevant information regarding events that justify the specific course of action. By selectively sharing only information that is relevant to team actions, situation understanding is leveraged without over-loading communications networks. This thesis describes the problem in detail, addresses its formalization, proposes an approach to cooperative situation assessment and presents results in two different domains, robots in a Urban Search and Rescue domain and patrol boats in a seacoast surveillance domain.