Evolutionary synthesis of dynamic motion and reconfiguration process for a modular robot M-TRAN

Abstract

In this paper we present a couple of evolutionary motion generation methods using genetic algorithms (GA) for self-reconfigurable modular robot M-TRAN and demonstrate their effectiveness through hardware experiments. Using these methods, feasible solutions with sufficient performance can be derived for a motion generation problem with high complexity coming from huge configuration and motion possibilities of the robot. The first method called ERSS (Evolutionary Reconfiguration Sequence Synthesis) applies GA (Genetic Algorithm) to evolution of motion sequence including configuration changes though natural genetic representation. The effectiveness of the generated full-body dynamic motions are verified through hardware experiments. The second method called ALPG (Automatic Locomotion Pattern Generation) Method seeks locomotion pattern using a neural oscillator as a CPG (Central Pattern Generator) model and GA to optimize the parameters for locomotion. A number of efficient locomotion patterns has been derived, which are also experimentally verified.