An Approach to Parallel Processing of Dynamic Robot Models

A new parallel-processing scheme for robot dynamics compu tation on a multiprocessor system is described. The model to be processed is the customized numeric-symbolic dynamic model developed in our previous works (1983; 1985a, b), where each element of the dynamic model matrices is com puted by an independent procedure (subtask). This feature is very convenient for multiprocessing since the precedence relations are reduced to a minimum. The parallel-processing scheme employs the distribution of subtasks among CPUs according to a modified branch-and-bound (BB) method combined with the largest-processing-time-first (LPTF) algo rithm. Further, this method is extended to cover the parallel computation of driving torques, once the dynamic model matrices have been computed. The method is extremely effi cient. The computational time for the Stanford manipulator was reduced 5.96 times by the use of six processors; i.e., the processors were active 99.4% of time. The programs to be executed on the processors with a given manipulator are computer generated, so no human intervention is needed to handle a new robot. The data transfer between the processors is almost negligible. The method is illustrated by two exam ples.