Automatic locomotion pattern generation for modular robots

Locomotion, one of the most basic robotic functions, has been widely studied for several types of robots. As for self-reconfigurable modular robots, there are two types of locomotion; one type is realized as a series of self-reconfiguration and the other is realized as a whole body motion such as walking and crawling. Even for the latter type of locomotion, designing control method is more difficult than ordinary robots. This is because the module configuration includes many degrees of freedom and there are a wide variety of possible configurations. We propose an offline method to generate a locomotion pattern automatically for a modular robot in an arbitrary module configuration, which utilizes a neural oscillator as a controller of the joint motor and evolutionary computation method for optimization of the neural oscillator network, which determines the performance of locomotion. We confirm the validity of the method by software simulation and hardware experiments.