Low-Level Cognitive Modeling of Aircrew Function Using the Soar Artificial Intelligence Architecture

摘要

Pilot vehicle interface (PVI) testing usually requires extensive human-in-the-loop (HITL) simulation. An alternative to HITL testing is to model human computer interaction with an automated cognitive engineering tool. This study used Soar cognitive modeling to compare the effectiveness of an existing and proposed PVI for air-to-ground Maverick missile missions. The baseline interface used a Forward-Looking Infrared Radar (FLIR) to detect and designate targets. The improved PVI had an enhanced FLIR and added Real-Time Information in the Cockpit (RTIC) with annotated overhead imagery of the target area. The Soar software architecture was chosen to model pilot cognition, although target acquisition was more dependent on the pilot's visual and motor functions than cognition. The Soar model accurately predicated faster target acquisition for the RTIC PVI and faster target acquisition for reduced scene complexity. Although not statistically significant, the Soar model correctly indicated that increased scene complexity caused larger increases in target acquisition time for the RTIC PVI condition as compared to the baseline condition (HITL 179% increase, Soar 47% increase). Furthermore, Soar was the only model that accurately predicted increased latency in the RTIC condition while both Cognitive and Traditional Task Analyses predicted decreased latencies.