Coevolving behavior and morphology of simple agents that model small-scale robots

Humanity have long strived to create microscopic machines for various purposes. Most prominent of them employ nano-robots for medical purposes and procedures, otherwise deemed hard or impossible to perform. However, the main advantage of this kind, of machines is also their main drawback - their small size. The miniature scale they work in, brings a lot of problems, such as not having enough space for the computational power needed for their operation, or the specifics of the laws of physic that govern their behavior. In our study we focus on the former challenge, by introducing a new standpoint to the well-studied predator-prey pursuit problem (PPPP) using an implementation of very simple predator agents, using nano-robots designed to be morphologically simple. They feature direct mapping of the (few) perceived environmental variables into corresponding pairs of rotational velocities of the wheels' motors. Our previous, unpublished work showed that the classic problem with agents that use straightforward sensor, do not yield favorable results as they solve only a few of the initial test situations. We implemented genetic algorithm to evolve such a mapping that results in an optimal successful behavioral of the team of predator agents. In addition, to cope with the previously described issue, we introduced a simple change to the agents in order to improve the generality of the evolved behavior for additional test situations. Our approach is to implement an angular offset to the visibility sensor beam relative to the longitudinal axis of the agents. We added the offset to the genetic algorithm in order to define the best possible value, that introduces most efficient and consistent solution results. The successfully evolved behavior can be used in nano-robots to deliver medicine, locate and destroy cancer cells, pinpoint microscopic imaging, etc.1