Evolving Aggregation Behaviors inSwarms fromanEvolutionary Algorithms Point ofView

Aggregation is a great example of self-organized behavior, largely observed in animal groups and social insects. It results in the most common social phenomena and is considered a key factor in favoring the emergence of collective decisions. This work analyzes the performance of different evolution strategies by considering collective decisions in a group of robots through aggregation behaviors. The analysis was conducted by comparing two evolutionary algorithms: the standard CMA-ES (Covariance Matrix Adaptation Evolution Strategy) and the xNES (Expo- nential Natural Evolution Strategy). The simulated scenarios include environments where the position and size of the aggregation sites (i.e., nests) are variable. We focus on aggregation performance to test whether the robots’ distribution reflects the rel- ative proportion of the site. Finally, we investigate how communication capabilities influence total performance. Results indicate that both algorithms successfully solve the evolutionary problem when robots are allowed to exchange information with other mates. Conversely, CMA-ES significantly outperforms xNES when agents cannot explicitly communicate. Moreover, the former method is qualitatively better at optimizing the level of aggregation.