An Implementation of a Parallel Algorithm for Real-Time Model- Based Control on a Network of Microprocessors

Abstract

Model-based trajectory control is a class of advanced manip ulator control strategies that are based on the kinematic and dynamic models of manipulators. Many schemes of model- based control have been proposed; however, most of them are limited within the stage of simulation study and very few have been installed on actual manipulators. The main reason is that the computations required for inverse dynamics are far beyond the ability of the present commercially available microprocessors. In order to achieve real-time high sample- rate control, we have proposed a parallel processing scheme for inverse dynamic computations that is called the resolved Newton-Euler algorithm. It resolves the process for computing the desired torques into subprocesses that have almost com plete concurrency. Because these subprocesses communicate with each other in a regular and local way, they can be di rectly mapped on a distributed processing network system. In this paper, we present a real-time parallel computation architecture for implementing the resolved Newton-Euler algorithm composed of a host personal computer and a net work of microprocessors and transputers. All programs are written in a high-level language, occam. This computation scheme is applied to the trajectory control of the PUMA 560 manipulator. Our controller achieves the sampling period of 0.66 ms.