Improving Survivability in Environment-driven Distributed Evolutionary Algorithms through Explicit Relative Fitness and Fitness Proportionate Communication

Abstract

Ensuring the integrity of a robot swarm in terms of maintaininga stable population of functioning robots over longperiods of time is a mandatory prerequisite for building morecomplex systems that achieve user-defined tasks. mEDEAis an environment-driven evolutionary algorithm that providespromising results using an implicit fitness functioncombined with a random genome selection operator. Motivatedby the need to sustain a large population with sufficientspare energy to carry out user-defined tasks in the future,we develop an explicit fitness metric providing a measureof fitness that is relative to surrounding robots andexamine two methods by which it can influence spread ofgenomes. Experimental results in simulation find that use ofthe fitness-function provides significant improvements overthe original algorithm; in particular, a method that influencesthe frequency and range of broadcasting when combinedwith random selection has the potential to conserveenergy whilst maintaining performance, a critical factor forphysical robots.