Evolving Aggregation Behavior for Robot Swarms: A Cost Analysis for Distinct Fitness Functions

Evolving behaviors for swarm robotic systems offers interesting emerged strategies which may be complex and unpredictable by an explicit behavioral controller design. However, even in the evolutionary case, there are critical choices regarding the design of the evolutionary algorithm that a roboticist should take into account to achieve desired goal with a reasonable efficiency. Among these design choices, adopting an appropriate fitness function is a crucial task, since it directly affects the resulting evolved strategy of a robot group. For the evolution of a single goal, different fitness functions can be used and their efficiencies can be compared. In this study, we chose complete aggregation as the desired goal for a robot swarm and compared the performances and costs of two distinct fitness functions in a simulated environment. Whilst the performance analysis consists of testing the average success rates, the cost analysis measures average time and distance taken by robots up to the successful formation. The results showed that for small communication ranges there is a trade-off between performance and cost in the fitness function selection; and hybrid control models can be utilized to overcome this issue to some extent