Distributed Cognitive Control System for a Humanoid Robot
Kazuhiko Kawamura, Richard Alan Peters II, Robert E. Bodenheimer, Nilanjan Sarkar, Juyi Park,
Charles A. Clifton, Albert W. Spratley, and Kimberly Hambuchen
Abstract
During the last decade, researchers at Vanderbilt have been developing a humanoid robot
called the Intelligent Soft Arm Control (ISAC). This paper describes ISAC in terms of its
software components and with respect to the design philosophy that has evolved over the
course of its development. Central to the control system is a parallel, distributed software
architecture, comprising a set of independent software ob jects or agents that execute as
needed on standard PCs linked via Ethernet. Fundamental to the design philosophy is
the direct physical interaction of the robot with people. Initially, this philosophy guided
application development. Yet over time it became apparent that such interaction may
be necessary for the acquisition of intelligent behaviors by an agent in a human-centered
environment. Concurrent to that evolution was a shift from a programmer's high-level
specification of action toward the robot's own motion acquisition of primitive behaviors
through sensory-motor coordination (SMC) and task learning through cognitive control
and working memory. Described is the parallel distributed cognitive control architecture
and the advantages and limitations that have guided its development. Primary structures
for sensing, memory, and cognition are described. Motion learning through teleoperation
and fault diagnosis through system health monitoring are also covered. The generality of
the control system is discussed in terms of its applicability to physically heterogeneous
robots and multi-robot systems.
Bobby Bodenheimer 