| Popis: |
Recent developments in the embodied approach to understanding the generation of adaptive behaviour, suggests that the design of adaptive neural circuits for rhythmic motor patterns should not be done in isolation from an appreciation, and indeed exploitation, of neural-body-environment interactions. Utilising spontaneous mutual entrainment between neural systems and physical bodies provides a useful passage to the regions of phase space which are naturally structured by the neuralbody- environmental interactions. A growing body of work has provided evidence that chaotic dynamics can be useful in allowing embodied systems to spontaneously explore potentially useful motor patterns. However, up until now there has been no general integrated neural system that allows goal-directed, online, realtime exploration and capture of motor patterns without recourse to external monitoring, evaluation or training methods. For the first time, we introduce such a system in the form of a fully dynamic neural system, exploiting intrinsic chaotic dynamics, for the exploration and learning of the possible locomotion patterns of an articulated robot of an arbitrary morphology in an unknown environment. The controller is modelled as a network of neural oscillators which are coupled only through physical embodiment, and goal directed exploration of coordinated motor patterns is achieved by a chaotic search using adaptive bifurcation. The phase space of the indirectly coupled neural-body-environment system contains multiple transient or permanent self-organised dynamics each of which is a candidate for a locomotion behaviour. The adaptive bifurcation enables the system orbit to wander through various phase-coordinated states using its intrinsic chaotic dynamics as a driving force and stabilises the system on to one of the states matching the given goal criteria. In order to improve the sustainability of useful transient patterns, sensory homeostasis has been introduced which results in an increased diversity of motor outputs, thus achieving multi-scale exploration. A rhythmic pattern discovered by this process is memorised and sustained by changing the wiring between initially disconnected oscillators using an adaptive synchronisation method. The dynamical nature of the weak coupling through physical embodiment allows this adaptive weight learning to be easily integrated, thus forming a continuous exploration-learning system. Our result shows that the novel neuro-robotic system is able to create and learn a number of emergent locomotion behaviours for a wide range of body configurations and physical environment, and can re-adapt after sustaining damage. The implications and analyses of these results for investigating the generality and limitations of the proposed system are discussed. |
