Neuroevolution of Sequential Behavior in Multi-Goal Navigation Task
Sergey Muratov, Konstantin Lakhman, Mikhail Lakhman · 2014
In this paper we consider a problem of evolutionary gener-ation of behavioral sequences for a mobile robot in the en-vironment with multiple goals. Many real-world tasks can be abstracted in terms of generation of actions sequences. To address this problem we utilize a neuroevolutionary al-gorithm, based on a neuron duplication, as a method to gen-erate the robot’s controller for multi-goal sequential naviga-tion. We show that neuroevolution produces stable behavioral strategies that can be successfully utilized even in the case of changing or randomized environment.