Neuron model interpretation of a cyclic motion control concept

Elastic properties of muscles and tendons are assumed to play a central role for the energy efficiency and robustness of locomotion in biological systems. Yet, the way in which the nervous system controls highly nonlinear body dynamics to produce stable periodic motions is far from being well understood. On the basis of a simple but very effective control law, which we developed and verified for variable impedance robots, we propose a controller model, which might be a very plausible hypothesis also for biological systems. The original robot controller has a bang-bang action triggered by the generalized force acting along a coordinate corresponding to the principal oscillation mode of the system. This coordinate is computed in a model-free, adaptive manner. It turns out that the control law can be easily realized with a neural network, whose weights are adapted according to the Hebbian learning rule. If this hypothesis is confirmed, cyclic body motions can be very easily and robustly implemented, with a surprisingly small number of neurons.