ABSTRACT
This is possible due to the central nervous system (CNS), which is equipped with a set of adaptive control mechanisms that keeps learning. One of the influential paradigms in understanding human motor control and its adaptation is the so-called force field paradigm [50]. In this paradigm, the human is asked to perform reaching movements while holding on to a handle. The handle is in fact a robot that can induce desired forces depending on the state of the subject’s hand. With this setup it has been shown that humans learn to adapt their movements apparently to counteract perturbations in a state-dependent way. Furthermore, it is shown that this is not achieved by stiffening of the muscles, but rather by active force generation to react to the learned force field. This indicates that the CNS optimizes energy by building internal predictive models to generate forces in an anticipation of the perturbation. The debate as to what is the exact mechanism of movement generation employed by the CNS is not settled. On one side, researchers propose that the observed effects can be explained by the optimal feedback control employed by the CNS. The other, more classical, view is that the desired trajectory governing reaching movements is tracked by learning internal models (see the next subsection) to cancel out the perturbations. Yet, another, rather reconciliatory view is, that which the CNS actually does, to reoptimize the desired trajectory in the face of large perturbations [26]. From a computational view, adoption of pure optimal force control (OFC) seems computationally infeasible for robotic systems, leaving the reoptimization view the most reasonable candidate for robotic implementations.
