Design and Analysis of a Swarm Robotics Algorithm

Abstract

Progressive minimal criteria novelty search can perform in a di erent way than objective-based search. It can reach better solutions than objective-based search for a multi-objective task for swarm robots. This work examined di erent changes in the evaluation module of the multi-objective genetic algorithm settings. These changes can enhance one objective, but another objective may get worse. This work proved that using harder problem for learning during the evolutionary algorithm will give us better solutions for easier problems. This work showed that a genetic algorithm tness evaluation module can be easily parallelized as the evaluation module is a group of individuals that should solve some problem independently. This makes 65 Chapter 6 the distribution process only handles network issues and does not need to work a lot on the correctness of the algorithm after distribution. At the start of the evaluation module, at each generation, the algorithm stops working and gives the control to the network layer which handles sending individuals to machines and receives penalty value, and after all individuals had been evaluated, the control returns to the genetic algorithm to complete normally.