Distributed real-time control of a spatial robot juggler

Abstract

The Yale spatial juggler and an emerging set of working principles for the design and implementation of embedded real-time distributed controllers are described. The robot uses a distributed network of transputers to process stereo camera data and control the torque of a three-degree-of-freedom arm to juggle a ball. The juggling algorithm is a direct extension of a novel class of nonlinear feedback controllers, called mirror laws. The algorithm takes the form of a mathematical expression that specifies robot position as a function of the ball's position and velocity. The programming approach, called geometric programming, substitutes event-driven dynamical processes and geometrical transformations for a more syntactically oriented if-then-else approach.