The paper reports on a discovery field exercise used to examine how disaster responders can use an audio and video equipped robot to interact with a trapped victim. In the exercise, a small robot with two-way video and audio communication was inserted into a physically simulated building collapse next to a trapped victim, and was provided to a team of trained responders as a means for performing remote triage and victim monitoring. The interaction between the responders and the victim was examined, with emphasis on how the responders adapted to different video and audio capabilities, and how they might have responded to different populations and injuries that may limit communication. The ad hoc interaction protocols used by the responders were observed in the field exercise, and four interaction schemes were identified: Two-way Video with Two-way Audio, One-way Video (from Robot to Responders) with Two-way Audio, Two-way Video with no Audio, and One-way Video (from Robot to Responders) with no audio. The interaction schemes are defined according to the minimum capabilities of the robot and victim, the requirements of the responders, and preliminary protocols required for each interaction scheme. From observations made about the exercise, the paper identifies minimalistic interfaces and transparency of robot state as key areas for improving a robotmediated interaction between responders and victims.