In this work we study a way of introducing temporal information in the structure of artificial neural networks that will be used as behavioral controllers for real mobile robots operating in unstructured environments. We introduce networks with delays in their synapses as the building block for these controllers and the evolutionary methodology employed for obtaining them in simulation. The effects of different types of noise added during evolution on the robustness of the controllers in the real robot are commented. Two examples of behaviors that will require time reasning in our robot implementation are presented: wall following and homing.
展开▼
