We have developed an execution-monitoring framework that can easily be adapted to produce interactive monitors for agent teams in dynamic domains. To support this claim, we describe two dynamic, data-rich, real-world domains and the Execution Assistants (EAs) we have implemented using our framework. Our first domain, Army small unit operations (SUO), has hundreds of mobile, geographically distributed agents, which are a combination of humans, robots, and vehicles. The other domain, UV-Robotics , is described in Section 7 and has teams composed of a handful of cooperating, unmanned ground and air vehicles (UGVs and UAVs) and a human controller. Both domains involve unpredictable adversaries in the vicinity of the team members.
We originally developed our monitoring framework for the SUO domain using several person-months of effort, although the majority of the effort was in knowledge acquisition and modeling. The SUO monitoring framework, described below, was designed to be modular and to support the easy insertion of domain-specific (and user-customized) system components, such as task models, monitoring algorithms, and value-of-information estimators. Our design was validated when we implemented a complex execution monitor in the UV-Robotics domain in about one person-week (as described in Section 7.2). The UV EA uses the same plan representation and basic architecture as the SUO EA, but the inputs are different as are the tasks and the monitoring algorithms that respond to the inputs and generate alerts.
The majority of our framework also applies to completely automated execution monitoring as demonstrated by the UV EA. A UV EA runs on each robot in the team and is used to autonomously adjust the robot control by blending desired behaviors and automatically revising plans during execution. The UV EA also provides alerts to any human controller who is monitoring the robots. While the framework described in this section is general, we follow it with some domain-specific details which clarify the concepts and tradeoffs. These details may not be of interest to all readers.