Speedup Improvement on Automatic Robot Programming by Parallel Genetic Programming
Shisanu Tongchim, Prabhas Chongstitvatana · 1999
Genetic Programming has been successfully used to perform automatic generation of robot programs. However, to improve robustness of the generated robot programs, each candidate solution was evaluated under many environments which required the substantial processing time. This study proposed a parallel implementation to reduce the execution time. By using a coarse-grained model for parallelization, called Island Model, a near linear speedup was achieved with small communication overhead. In addition, the barrier synchronization was identified to be the primary source of the overhead.