AutoMoDe-Arlequin: Neural Networks as Behavioral Modules for the Automatic Design of Probabilistic Finite-State Machines

Abstract

We present Arlequin, an off-line automatic design method that produces control software for robot swarms by combining behavioral neural-network modules generated via neuro-evolution. The neural-network modules are automatically generated once, in a mission-agnostic way, and are then automatically assembled into probabilistic finite-state machines to perform various missions. With Arlequin, our goal is to reduce the amount of human intervention that is required for the implementation or the operation of previously published modular design methods. Simultaneously, we assess whether neuro-evolution can be used in a modular design method to produce control software that crosses the reality gap satisfactorily. We present robot experiments in which we compare Arlequin with Chocolate, a state of the art modular design method, and EvoStick, a traditional neuro-evolutionary swarm robotics method. The preliminary results suggest that automatically combining neural-network modules into probabilistic finite-state machines is a promising approach to the automatic conception of control software for robot swarms.