Combining Conflicting Environmental and Task Requirements in Evolutionary Robotics

Abstract

The MONEE framework endows collective adaptive robotic systems with the ability to combine environment- and task-driven selection pressures: it enables distributed online algorithms for learning behaviours that ensure both survival and accomplishment of user-defined tasks. This paper explores the trade-off between these two requirements that evolution must establish when the task is detrimental to survival. To this end, we investigate experiments with populations of 100 simulated robots in a foraging task scenario where successfully collecting resources negatively impacts an individual's remaining lifetime. We find that the population remains effective at the task of collecting pucks even when the negative impact of collecting a puck is as bad as halving the remaining lifetime. A quantitative analysis of the selection pressures reveals that the task-based selection exerts a higher pressure than the environment.