Abstract
In this report we present some initial results of our work completed thus far on "Computational Structures for Robot Control". A SIMD architecture with the crossbar interprocessor network which achieves the parallel processing execution time lower bound of o( [a1n ]), where a1 is a constant and n is the number of manipulator joints, for the computation of the inverse dynamics problem, is discussed. A novel SIMD task scheduling algorithm that optimizes the parallel processing performance on the indicated architecture is also delineated. Simulations performed on this architecture show speedup factor of 3.4 over previous related work completed for the evaluation of the specified problem, is achieved. Parallel processing of PUMA forward and inverse kinematics solutions is next investigated using a particular scheduling algorithm. In addition, a custom bit-serial array architecture is designed for the computation of the inverse dynamics problem within the bit-serial execution time lower bound of o(c1k + c2kn), where c1 and c2 are specified constants, k is the word length, and n is the number of manipulator joints. Finally, mapping of the Newton-Euler equations onto a fixed systolic array is investigated. A balanced architecture for the inverse dynamics problem which achieves the systolic execution time lower bound for the specified problem is depicted. Please note again that these results are only preliminary and improvements to our algorithms and architectures are currently still being made.
Date of this Version
10-1-1987