A team of researchers from the Naval Research Laboratory received the Best Paper Award at FLAIRS-24 (the 24th Florida Artificial Intelligence Research Society Conference), sponsored by the Association for the Advancement of Artificial Intelligence. The paper, Active and Interactive Learning of Goal Selection Knowledge, describes NRL's research on goal-driven autonomy.
Goal-driven autonomy, or GDA, is a meta-reasoning model of high-level autonomy that permits agents to self-select their goals. The researchers' paper describes algorithms in which a GDA agent dynamically interacts with a subject matter expert to acquire and confirm knowledge that maps goals to be formulated in response to encountering different types of operational situations.
The researchers focused their study on demonstrating the utility of this approach in the context of scenarios defined using the Tactical Action Officer (TAO) Sandbox, a training simulator used at the Surface Warfare Officers School. Future objectives for the research include assessing GDA approaches for controlling unmanned underwater and unmanned sea surface vehicles.
The authors on the paper were Jay Powell, NRL's Summer Employment Program; Matt Molineaux, Knexus Research Corporation; and David Aha, NRL's Information Technology Division.
This collaboration was supported by a project, titled Evaluation Support and Event Discovery for Case-Based Active Transfer Learning, that is sponsored by ONR as part of the Active Transfer Learning Program. Powell collaborated on this project through NRL's Summer Employment Program during the summer of 2010 and has recently received his doctorate from Indiana University.