Servo - July 2008

GEERHEAD

Photographers try to avoid empty space in their photos. This helps ensure that photos contain as much relevant visual information as possible. Lewis weighs the rule of thirds and the rule around empty space one against the other whenever they conflict, to take the best pictures.

Lewis can also think for himself when taking photos. He is free to break the rules altogether and take feedback about his images. He uses this information to learn which rules to break and when in order to deliver great photos based on a sort of photographic instinct.

Sharing a more whimsical moment with Lewis are members of the Media and Machines Lab (from left): Assistant Professor William Smart; Assistant Professor Cindy Grimm (seated): two of the lab’s founders, Shannon Lieberg, Engineering Class of ‘04 and research assistant Michael Dixon, B.S./M.A. ‘03; and Nik Melchior, a fifth year B.S./M.S. student in computer science and engineering. Members are showing off the “playful props” used by Lewis in the lab.

Live Test

Researchers tested the Lewis robot photographer on a group of 5,000 subjects over a period of 40 hours. Lewis took 3,000 pictures in that period. During this 40 hour run, people (guests at a large technology event) either ignored the robot completely or tried to interact with it.

Because the robot wasn’t instilled with the ability to interact, people quickly dispensed with it and began socializing with other people in the crowd. Because people ignored the robot, they relaxed and acted naturally, enabling the robot to take candid, natural pictures.

it there for? In this version of the robot it made a noise, sounding an alarm or signal when it had taken a picture. However, the sound wasn’t loud enough for most people to hear.

If the signal were louder, this would communicate to people in the robot’s proximity that it had just taken a picture. This would form some level of communication between the robot and those people, and provide some simple basis for interaction.

The robot had no way of telling people to hold still or say cheese. It took four seconds for the robot to line up shots, in which time people might

hold still to get their picture taken, or they might move around. After the robot’s test run, people suggested that the robot actually say cheese or show a picture of a “birdie” (as in, look at the birdie) to signal that it was about to take a picture.

People waiting in front of the robot hoping to have their pictures taken were often disappointed when the robot was navigating, getting its bearings, or homing in on a landmark instead of taking pictures at that particular moment. However, the

Lewis close-up head shot with camera.

Results

Among other things, researchers determined that the robot should have a sort of bi-modal capability. If someone is trying to interact with it, it should stop what it is doing and interact with those people, taking their pictures where possible. If no one is trying to interact with the robot, it should blend into the background and continue to take candid shots. This version of the robot is only capable of blending in. So, the robot will ignore people who want to interact with it or who specifically look to have their picture taken.