Motor adaptation as an optimal combination of computational strategies

Abstract
To efficiently and accurately manipulate objects, the nervous system must adjust motor commands based on experience. Four major adaptive strategies that could help achieve this goal are: internal model formation of the environmental dynamics, minimizing force, trajectory planning, and selectively stiffening the arm. We measured motor adaptation to a robotic force field with and without a large background force requirement. We then developed a computational model of motor adaptation that allowed the relative contribution of the four strategies to be estimated. Motor adaptation was best modeled as a blend of strategies, with internal model formation playing a greater role when forces were smaller and predictable; impedance control had a higher priority when forces were smaller and unpredictable; force minimization was more important when forces were larger; and trajectory planning was involved in both large and small background force conditions. These results are consistent with the viewpoint that the nervous system effectively seeks to minimize a cost-function containing force, stiffness, and position error terms.