A heterogeneous multiprocessor programming method with application to robotic control systems

The testing of robotic control theory on robots has resulted in the design of robot control systems which are multiprocessor based. The authors describe a method by which a control system is implemented on a heterogeneous multiprocessor. The method has three parts: decomposition of the control algorithm into concurrent units of computation; converting the computational units into software tasks; and the allocation of the tasks to processors. The allocation procedure is recognized as a NP-complete problem. By using an appropriate model for the multiprocessor, it is possible to extend the method used to solve the allocation problem for homogeneous processors to the heterogeneous case. Extensions to the algorithms which calculate lower bounds on the completion time of partial allocations are proposed. These lower bounds are used in a branch and bound (BB) technique that has been shown to be effective at solving the allocation problem for homogeneous processors. >