Parallel distributed architecture for motor control

Redundancy and task variety are among the main sources of complexity in motor control. Muscle elasticity provides the starting point for an answer. The spring-like properties of muscles make it possible to define precisely the notion of posture of the whole body: a configuration of minimum potential energy, coded by the set of muscle elastic paramenters. The body is attracted by postures, whether static or dynamic. Differently from the Hopfield nets or the Boltzman machine, the muscle potential energy is a real physical entity and not an abstract analogy. However, what the motor control system needs to exploit the underlying physics, is an analogy of muscle elasticity, i.e. an internal model that can be used for task formation. We formulated such an analogy as a parallel distributed architecture that is able to generate attracting postures that fit the specific requirements of a task. The attracting postures serve two purposes: an outflow of modifications of all the muscle parameters and an inflow of expected trajectories of all the body parts (an efference copy) to be compared with the actual sensory inflow for monitoring the smooth development of an action.