Neuro-evolved Agent-based Cooperative Controller for a Behavior- based Autonomous Robot

-we present in this work a cooperative control paradigm for the autonomous navigation of a mobile robot and demonstrate that the cooperative controller learns faster and better than a centralized one. Behaviors emerge from the neuro-evolved controller, in order to achieve a designed task and without been defined at design stage. In our proposal, a robotic agent is divided into sub-agents, each one controlling one sensor or actuator element of the robot. Meanwhile the sub-agent learns to handle the element, it also learns to cooperate with the other ones. The emergence of behaviors happens when the coevolution of several sub-agents embodied into the single robotic agent stabilizes. A distributed version of the ESP neuro-evolving algorithm is used for the evolution of the overall distributed controller.