Decision Support for Anomaly Detection and Recovery for Unmanned System (ADRUS)

Unmanned systems have proven their value in combat operations by delivering unprecedented mission performance. Unfortunately, the current unmanned systems are predominantly tele-operated, which tie up many skilled operators per unmanned system. Thus, increasing demands for unmanned system cannot be met with the current state-of-the-art. The problem of low levels of autonomy is further exacerbated by the lack of decision support for behavioral anomaly detection and subsequent recovery planning. We will address this capability gap by developing approaches for increasing the level of autonomy from tele-operated to human supervised as follows. In particular, we will develop ADRUS, a decision support system for anomaly detection and recovery for unmanned system for multi-vehicle missions. ADRUS will provide automated monitoring, perform continuous anomaly detection and analysis in the mission context, analyze root causes for the anomaly and explain its findings to the mission personnel. It will go a step further and recommend plans to recover from the anomaly to minimize disruptions and maximize mission success. To develop these capabilities, we will investigate the use of a variety of probabilistic causal models that exploit the knowledge of mission to assess the deviations and provide accurate alerts. We will investigate fast and incremental automated planning approaches that exploit current resource knowledge to compute effective recovery plans. In developing ADRUS, we will consider human factor issues, such as reduction of cognitive load by developing appropriate alert presentation techniques and human-machine trust by developing decision explanation and justification abilities. We will demonstrate ADRUS feasibility by developing a prototype reference implementation and evaluating it using multi-vehicle mission scenarios.