The authors of this paper used a four-foot tall robot to investigate areas of autonomous navigation and robot-human interaction. They programmed into the robot the “artistic” definition of what defines a “good” photograph. Subjects were identified by face detection, based on skin color and location on the object. The decision to take a picture was then determined by distance from the subject, reachability (whether or not the robot could get to the shooting distance), and arrangement of the subjects (whether or not there were overlapping subjects, and how the subjects were arranged in the frame).
An “ideal” photograph was defined by these basic photographic composition rules: the rule of thirds, the empty space rule, the no middle rule, and the not on the edge rule. Variations were programmed for pictures of single subjects versus group shots.
The robot was tested first at two professional conferences, and later at a wedding and a birthday party. The robot took several thousand pictures over five days at the first conference, and several hundred shots at the other events. In all, about one-third of the pictures were usable. The paper contains some of the images, and discusses lessons learned. The authors learned that the biggest challenge is in the navigation of the environment of the event at which the robot operates. They suggest that future systems of this kind be highly modular, so that they can be customized. “Even the same physical location changes from deployment to deployment, making it necessary to adapt the solution every time it is deployed” (page 46).