The emergence of communication through evolutionary computation in a swarm of initially non-communicative robots is a highly complex research topic that has vastly captured the attention in the swarm robotics field.In this paper, we empirically study the emergence of communication as a result of an evolutionary algorithm in a swarm of simulated robots with the objective of solving an orientation consensus problem. Specifically, the consensus is reached provided that the heading orientations of the robots point into the same direction. The robots are controlled by Continuous-Time Recurrent Neural Networks whose parameters are evolved using a genetic algorithm. Once evolution is concluded, we assess the performance and scalability of the swarm behavior and the type of communication that emerged. The study is accomplished by means of an statistical analysis of the communication variables produced in a sample of 50 independent simulations. The conducted analysis suggests that the emerged communication is situated, meaning that both the message content and its associated context about the environment are informative and useful in the communication. Very interestingly, the environment context is the only piece of information actually relevant for reaching the consensus. On the contrary, the abstract message content is crucial for drastically reducing the rotation speed of the robots after the orientation consensus is achieved. |
*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.