Chaotic Search of Emergent Locomotion Patterns for a Bodily Coupled Robotic System

We study a novel deterministic online process for the exploration and capture of possible locomotion patterns of a simulated articulated robot with an arbitrary morphology in an unknown physical environment. The robot controller is modelled as a network of neural oscillators which are coupled indirectly through physical embodiment. Goal directed exploration of coordinated motor patterns is achieved by a chaotic search method using adaptive bifurcation. The phase space of the indirectly coupled neural-body-environment system contains multiple phase-locked states each of which is a candidate for driving efficient locomotion. By varying the chaoticity of the system as a function of evaluation signal, it is able to chaotically wander through various phase-locked states and stabilise on one of the states matching the given criteria. The nature of the weak coupling through physical embodiment ensures that only physically stable locomotion patterns emerge as coherent states, which implies the emergent pattern is well suited for open-loop control with little or no sensory inputs.