The biological motor control system is an adaptive system exhibiting vast redundancy at all its hierarchical levels. Redundancy improves reliability and flexibility and might be the salient reason for the superb dexterity of human motor control. However, introducing redundancy in an inversely controlled object results in an ill-posed problem.

Most previous research addressed this issue using a pseudo-inverse or other related criteria to find a single solution. The recent progress in computer technology, and in the field of neurocomputation, enables us to consider an architecture that finds and retains all the solutions, and chooses one of them in real time according to criteria that might be altered under different circumstances.

We describe a general two-way model for redundancy control. This model includes the selection of a single solution by the dynamics of the system at the lower level, and a multiple controller at the higher level. We demonstrate the role of the system dynamics in facilitating the stereotypical features of rapid movements, and suggest an architecture as well as a theoretical framework for many-to-one function approximation and inversion. This architecture is a modified mixture of experts architecture named the Polyhedral Mixture of Linear Experts (PMLE). We show that the PMLE can be used in order to construct a multiple controller.

We hope that the tools and insights provided in this work will assist physiologists in modeling the biological motor control system, and engineers in exploiting the virtue of redundancy in artificial systems.

* This study was done at the Department of Electrical Engineering, Technion Israel Institute of Technology, under the supervision of Prof. Gideon Inbar and Dr. Ron Meir.