Neural computing for the coordination of pseudo-muscular actuators

Redundancy and concurrency in motor control are basic problems for multi-joint, multi-limb robots. The viscous-electric properties of pseudo-muscular actuators provide an organising factor because they define equilibrium postures that influence body movements as ´postural attractors´. The authors report the state of advancement of a pseudo-muscular actuator which uses polymer gels and describe a computational architecture which naturally exploits the actuator-mechanical structure. The main idea is that the mechanism of generation of synergies of motor commands can be formulated as the internal simulation of passive movement or (passive motion paradigm). They show how this approach can be formulated in connectionist terms as a multiple constraints satisfaction network, driven by a potential function similar to a Hopfield network. They call this model ´motor relaxation network´ (or M-net).<>