Execution monitoring in multi-agent environments

Agents in dynamic multi-agent environments must monitor their peers and the environment to execute individual and group plans, to ascertain their progress, and to detect failures. In practice, however, agents cannot continuously monitor all surroundings and their peers. This leads to uncertainty about monitored agents' states, and aggravates computational requirements. A key open question is thus how to limit monitoring activities while providing effective monitoring: The Monitoring Selectivity Problem. We investigate this question in the context of monitoring in teams of cooperation agents, in three complex, dynamic multi-agent domains, and in service of different monitoring tasks: Monitoring for coordination and teamwork failures, and monitoring distributed teams via their communications. We provide empirical and analytical answers to the monitoring selectivity problem, via Socially-Attentive Monitoring, which focuses on using knowledge about the relationships between monitored agents and the procedures used to maintain these relationships. We explore a family of socially attentive teamwork failure-detection algorithms under varying conditions of task distribution and uncertainty. We show that a centralized scheme using a complex algorithm trades correctness for completeness and requires monitoring all teammates. In contrast, a simple distributed teamwork-monitoring algorithm exploits agents' local state and results in correct and complete detection of teamwork failures, despite relying on limited, uncertain knowledge, and monitoring only key agents in a team. In monitoring a distributed team, we present heuristics for talking monitoring uncertainty (which results from the limited overheard communications), and provide empirical results demonstrating that socially attentive techniques can significantly reduce the uncertainty in such monitoring. Furthermore, we explore monitoring algorithms which trade-off efficiency for expressivity's, resulting in a limited-expressivity's algorithm that can detect failures and provide high-accuracy monitoring of a team and its members, using a single, constant-space structure. In addition, we report on the design, constant space, structure. In addition, we report on the design of a socially attentive monitoring system and demonstrate its generality in monitoring several coordination relationships in providing quantitative teamwork evaluation, and in diagnosing detected failures.