Effects of spike-frequency adaptation on neural models, with applications to biologically inspired robotics

Animals are impressive biological machines, and their ability to handle unstructured environments is something roboticists wish to emulate. The behavioural competence of animals derives largely from the functioning of their nervous systems. Mathematical modelling of the functioning of neurons may enable us to extract useful principles from biology to be applied in robotics. Here, several systems with relevance to biologically inspired robotics are analyzed. The qualitative dynamics of a biological property called spike-frequency adaptation are added to existing analog neural models, and analysis shows the conditions under which the augmented model can generate oscillatory solutions. A network of these augmented analog neurons is then used to generate a walking gait for a six-legged robot in such a way that the system recovers rapidly from perturbations to the legs. The dynamics of oscillations arising in two coupled populations of integrate-and-fire neurons are studied; an analysis of the system provides good predictions of the oscillatory period and the range of coupling strengths for which oscillations will occur. A signal-processing phenomenon known as noise-shaping, wherein noise in a system is shifted out of the low frequencies up into higher frequency ranges, is demonstrated in networks of integrate-and-fire and conductance-based neurons; it is shown that spike-frequency adaptation provides certain signal-processing advantages in such networks. The effect of spike-frequency adaptation on the variability in integrate-and-fire neurons' firing records is analyzed.