Computational Cognitive Modeling of Human-Robot Interaction Using a GOMS Methodology

The goal of this study was to use computational cognitive modeling to further understand human behavior and strategy in robotic rover control. To this end, GOMS (Goals, Operators, Methods, Selection Rules) Language models of rover control were constructed based on a task analysis and observations during human rover control trials. For the first model, we hypothesized control would be characterized by actions to prevent deviations from exact path following. The second model was developed based on an alternate hypothesis that operators commanded ballistic rover movements to approximate path direction. In manual trials, an operator was required to navigate a commercially available micro-rover along a defined path using a computer interface (providing remote environment information through a camera view) located in a room separate from the rover. The computational cognitive model was executed with a pseudo system interface (Java device) in real-time. Time-to-navigation completion and path tracking accuracy were recorded during the human and cognitive model trials with navigation circumstances being identical. Comparison of the GOMSL model outputs with human performance demonstrated the first model to be more precise than actual human control, but at the cost of time. The second model with the new navigation criteria appeared to be more plausible for representing operator behavior; however, model navigation times were consistently longer than the human. This was attributable to limitations of the modeling approach in representing human parallel processing and continuous control. Computational GOMS modeling approaches appear to have potential for describing interactive closed-loop rover control with continuous monitoring of feedback and corresponding control actions. Humans exhibit satisficing behavior in terms of rover direction and movement control versus minimizing errors from optimal navigation performance. Certain GOMSL modeling issues exist for applications to human-robot interaction and this research provides a first empirical insight.