Abstract
This paper presents a new task-generation method for robot-arm control computation on a multi-processor system with an arbitrary number of processors. In order to execute the effective parallel processing which achieves high speed-up ratio, the definition of optimal task-set is an essential problem. The proposed task-generation method, which is based on algebraic computation techniques, consists of three phases: 1) Derive a robot-control law by using symbolic languages, such as Mathematica or REDUCE, as a set of expressions; 2) Transform the expressions into a simplest possible form which does not contain any redundant operations. In this phase, the factorization algorithm which was recently proposed by the authors is used; 3) Divide each equation into subexpressions repeatedly to induce the parallelism of computation. Consequently, each of the optimized expressions is regarded as one task. By using the proposed task-generation method, an automatic parallelizing compiler is also developed. Experiment on the inverse dynamics computation of a six-joint robot-arm demonstrates the effectiveness and the usefulness of the proposed task-generation method.
