A Layered Architecture for Lifelike Robotic Motion

In this paper we present an architecture for the synthesis of realistic and emotionally expressive robotic motion. Our work borrows heavily from the computer graphics and classical animation communities which emphasize perceptual realism of motion, rather than mechanical efficiency. Principles of traditional animation, as articulated over the last century, are algorithmically embodied in our work, serving to create a sense of character and presence in a robot. Our architecture represents motion as layers of periodic function primitives. These primitives form more complex behaviors which are combined and sequenced by the Motion Compositor, a module that is parameterized by control weights determined by higher level abstractions. A signal-processing module modifies the overall synthesized motion, providing a complementary method for global control of affect. Our layered motion architecture is currently being used to explore the emerging field of personal robotics. In particular, we are experimenting with emotional communication between a physically embodied robot and humans. We have built a small pet-like robot which we operate via high level remote control to perform experiments in emotional interaction.