ocial Co on for ecove in lb 4 wings

In dynamic multi-agent environments, agents, often based on hand-crafted reactive plans (operators), form teams to collaborate in achieving joint goalsl. The complexity and unpredictability of such environments present the agents with countless opportunities for failures, which could not have been foreseen by the designer (e.g., due to incomplete task descriptions, unreliable sensors, etc.). Without additional information about their situation, and the ideal behavior expected of them, agents are unable to detect these failures or take corrective action. Previous approaches have focused on specifying explicit executionmonitoring conditions, but this is difficult in such environments, since it is like specifying the correct behavior -in the type of environment where the latter is problematic, the former may be just as hard. Instead, we propose a novel approach to failure detection, based on ideas from Social Comparison Theory. Here, agents use their team-mates as sources of information on the situation and the ideal behavior. The agents compare their own goals and plans to those of the other team members, in order to detect failures and correct their behavior. Since continuous communications from team-mates is usually costly, impractical, or erroneous, an agent can utilize its teamtracking capabilities to infer its comrades’ goals and plans. behavior. The modeling process uses the agent’s own reactive operators to also describe the goals of the other agents. The operators may be individual (specific to one agent) or joint (shared by a team). By contrasting its own operator hierarchy with those of the team members, the agent can identify discrepancies between the hierarchical reactive operators which it is executing and those of its team-mates. These differences raise alarm levels to indicate possible failures. These, in turn, are reasoned about explicitly to verify that indeed a failure has occurred, and then take corrective action. In the earlier example, our agent discovers it is stuck in place while the others are flying away by comparing its chosen method of flight to that of its comrades. In the second example, it discovers that while it is executing the joint reactive operator for formation-flying, other team-members are executing a different joint operator, which involves landing.