Development of a Recognition Primed Decision Agent for supervisory control of autonomy

Abstract

Unmanned Systems (UxVs) are becoming increasingly prevalent across both the Department of Defense (DoD) and commercial sectors. As automation becomes increasingly robust and systems transition from being merely automated to being more autonomous, there is a need for Human Machine Interfaces (HMI) to enable effective supervisory control of these systems. An ontology-driven decision support system has been designed to emulate Recognition-Primed Decision Making (RPD) that is exhibited in experts. This RPD Agent (RPDA) fuses real-time sensor data and vehicle telemetry with static rulesets based on tacit domain knowledge elicited from experts to generate Courses of Action (COA) that a user can pick from. By querying ontologies of human expert knowledge, real-time world states can be rapidly compared to instantiated rulesets to provide relatively novice users with extended domain expertise. By presenting the operator with a list of logically permissible COAs rather than generating the COAs themselves, one can employ effective supervisory control over multi-domain, multi-mission autonomy. This paper discusses the conceptual and functional design of the RPDA, the development of component domain ontologies to power the agent, conclusions, and future work to be performed.