Iñaki Fernández Pérez, PhD student in team Larsen will defend his thesis on December 19th at 13:30 in room A008.
The title is : “Distributed Embodied Evolutionary Adaptation of Behaviors in Swarms of Robotic Agents”.
Designing behaviors for swarms (large groups) of robots is a difficult challenge, which is accentuated when dealing with unknown and possibly changing environments. Consequently, automated approaches to design or learn such behaviors are of utmost
interest. Here, we present our contributions to the adaptation of behaviors in swarms of robotic agents using distributed Embodied Evolution (EE) algorithms. In these approaches, each robot in the swarm runs a learning algorithm, and robots exchange information when meeting to help each other learn. We present four main contributions that help understanding the dynamics of such distributed approaches. First, we evaluate the influence of the selection pressure in distributed EE.
Second, we evolve collaborative swarm behaviors, which is especially challenging in distributed settings. Third, we propose and validate a mechanism to evolve the structure of the robot controllers in a fully decentralized manner. Finally, we propose and provide preliminary results of an evolutionary algorithm that allows to retain skills learned in the past.