Using Coherence Inspectors To Determine The Reliability of Debriefed Epistemic Agents from Unmonitored Autonomous Team Missions

摘要

It is sometimes necessary to deploy autonomous multi-agent teams to perform analysis, interpretation, threat or safety assessments in environments that are either too remote, hazardous, or physically impractical to send human beings. It may also be the case that synchronous data communications along with any real-time video or audio monitoring of the team may not be possible or at the very least not dependable. In these situations if we are able to debrief the team at some point how do we rely on any analysis, or assessment that the team makes if we could not monitor its performance? On teams that perform assessment or analysis there is always the possibility that one or more members may make misidentifications, or false positives. Without the ability to monitor, it may not be clear that correct locations were visited or substances analyzed and assessed. It is also possible that team concluded its analysis and assessment without considering the entire environment or partial failures of team members in terms of inter-team communication or subtasks completion. In this paper we describe FACT, an experimental heterogeneous multi-agent system consisting of software and autonomous robots agents, and a Coherence Inspector (CI). The CI consist of a knowledge structure and a knowledge engineering technique used to determine the reliability of any analysis, interpretation, or assessment that FACT makes during the debriefing process.