Servo - July 2008

GEERHEAD

Here, Nik Melchior (left), a fifth year B.S./M.S. student in computer science and engineering, helps create the programming framework that allows others to command Lewis. Shannon Lieberg (center) of the Engineering class of ‘04, works Lewis’ controls with Assistant Professor Bill Smart.

Full frontal view of Lewis.

robot was programmed to take frequent pictures of human faces even when it may not have had the opportunity to focus in for a good picture. To better interact with subjects, the robot should have the ability to communicate which mode or “state” it is in to the subjects. So, if the robot is available to interact, it can communicate that, and so on.

Photographic subjects expected the robot to respond when waved at, like a human being would. However, the robot didn’t have this capacity either. When the robot did seem to react — because it turned toward someone by sheer coincidence when someone had waved at it, for example — people thought this meant it was more intelligent than it actually was.

In particular, when the camera pointed in their direction coincidentally in response to trying to hail the robot, this was mistaken for eye contact.

Lewis seemed intelligent to people when he did what appeared to be a double take. Because the robot face detection code was not optimized, the camera panned past the faces and beyond by the time the software determined it had detected a face. The camera then returned to focus on the face for the picture. This apparent “double take” humanized the robot in the eyes of on-lookers, attracting people to interact with the robot.

Likewise, other robot behaviors made the robot appear not so smart, even though these behaviors were quite intelligent for a robot in what they accomplished. One such behavior was looking at the wall (pointing its camera toward the wall) or moving along a wall in order to aid in its navigation. While the robot was trying to get its bearings, it appeared not to “see” anyone around it, and so looked dumb.

Conclusion

Continued research based on Lewis should address whether Lewis truly functions in two separate modes, whether the level of sophistication of the people interacting with Lewis has an impact, and whether the robot or the people around it should drive its interaction. SV