Application of parallel computing to robot dynamics

Abstract

In this paper an approach for the application of parallel processing to the dynamic analysis of robots based on the multibody system method is presented. The inherent structure of the symbolic equations of motion is used for partitioning those into independent modules for concurrent evaluation. The applied strategies for parallelization include the parallel evaluation of subsystem equations and the parallel computation of the inertia matrix along with its factorization, and of the force vectors and control inputs. The implementation of the parallel structures is discussed with respect to optimal utilization of transputer resources. The presented approach yields a strong reduction of computing time and supports real-time simulations of controlled robots, which is also shown in an example.