Parallel algorithms for dynamics of robotic manipulators
Date
1993
Authors
Pond, Christopher Burke
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The advent of relatively inexpensive parallel computing systems has motivated the investigation of parallel algorithms and architectures as a means of achieving more efficient robot dynamics algorithms. In this thesis, both the inverse dynamics and the forward (simulat10n) dynamics of rigid body manipulators are considered The parallelism inherent in six inverse dynamics formulations is analysed to determine the most efficient algorithm, both theoretically and on a network of transputers. An extensive comparison of the parallel performance of the algorithms is made by incorporating a hardware model of the transputer into a scheduling algorithm which searches for the best assignment of tasks to processors. Thus, new results comparing the performance of the algorithms including communication costs are presented. Although the logarithmic Recursive Newton-Euler algorithm for inverse dynamics is theoretically the fastest, its performance is constrained by the need for more tasks and more communications. The Resolved Newton-Euler algorithm is shown to execute the fastest on a transputer network, a result which is supported by analysis. A recently proposed macroparallel simulation dynamics algorithm is implemented as a proof-of-concept. The analysis of this algorithm concentrates on parallel performance and comparison with its serial implementat10n Performance models are partly based on experimental measurements rather than theoretical predictions and are used to assess the effects of serial parts of the algorithm on speedup and total execution time. Results indicate that minor serial computations in the algorithm can have a significant effect on the overall parallel performance.
Description
Keywords
UN SDG 9: Industry, Innovation, and Infrastructure