Virtual Spatiality in Agent Controllers: Encoding Compartmentalization

Applying methods of artificial evolution to synthesize robot controllers for complex tasks is still a challenging endeavor. We report an approach which might have the potential to improve the performance of evolutionary algorithms in the context of evolutionary robotics. We apply a controller concept that is inspired by signaling networks found in nature. The implementation of spatial features is based on Voronoi diagrams that describe a compartmentalization of the agent’s inner body. These compartments establish a virtual embodiment, including sensors and actuators, and influence the dynamics of virtual hormones. We report results for an exploring task and an object discrimination task. These results indicate that the controller, that determines the principle hormone dynamics, can successfully be evolved in parallel with the compartmentalizations, that determine the spatial features of the sensors, actuators, and hormones.