Methodology for study of human-robot social interaction in dangerous situations

Applications of robotics in dangerous domains such as search and rescue require new methodology for study of humanrobot interaction. Perceived danger evokes unique human psycho-physiological factors that influence perception, cognition and behavior. Human first responders are trained for victim psychology. Apart from real-life instances of disasters, studies of robots in this environment are difficult to perform safely and systematically with sufficient controls, fidelity, and in a manner that permits exact replication. Consequently, the trend to deploy rescue robots, for example, is proceeding largely without benefit of knowing whether human victims will readily cooperate with robot rescuers. The capability to deal with unique victim psychology has not been a testable requirement. We report on the methodology of an on-going study that uses virtual reality to provide a feature-rich immersive environment that is sufficient to evoke fear-related psychological response, provides simulation capability for robots, and enables systematic study trials with automated data collection via an embedded scripting language. The methodology presented provides an effective way to study human interaction with intelligent agents embodied as robots in application domains that would otherwise be impossible in the real world.