Calculating manipulator trajectories with parallel evolutionary strategies

Abstract

Due to the complexity of the search space, the calculation of robot trajectories is a well suited field for evolutionary algorithms. Owing to the time-expensive calculation of the evolutionary algorithm and especially the collision-detection algorithm, the calculation of robot trajectories is mostly not in real time. Therefore, this paper presents an approach for shortening the calculation time by parallelising the whole algorithm. First, the problem to be solved is described. Here, the special coding of the robot trajectories is presented. Then the parallel algorithm and its realisation on a cluster is introduced. Finally, the performance comparisons are presented.