Parallel computation of an algorithm in robotic control

The advantages of general purpose microprocessors for implementing digital control systems were recognised as soon as these devices appeared, but as the computational power of the microprocessor increased, so did the complexity of the control laws to be computed. Thus, despite the vastly increased capability of present day microprocessors, control engineers still wish for more speed. The basic problem has not changed. It is to compute the whole of a given control algorithm within the loop sample time, which is fixed by the dynamics of the physical process in question. If this cannot be done, there are two options: reduce the complexity of the algorithm, recognising that the system performance may suffer as a consequence; or use a device other than a general purpose microprocessor, accepting the increased hardware and software development costs that this implies. The purpose of this paper is to discuss a third option, that of parallel processing. Here the authors seek to reduce the computational period by distributing the algorithm over a number of cooperating general purpose processors. In particular, the device to be considered is the INMOS transputer and its associated programming language, occam