Humans move with a purpose. Thus the study of motor systems requires more than just the description of movement control. It demands a broad perspective based on anatomical, functional, and computational principles to describe how the nervous system can enable goal-directed behavior. The challenge in solving this problem remains daunting because the intrinsic neural codes that are used to generate motor actions are not fully known, the principles that lead to optimal performance are only just beginning to be modeled, and the system is nonstationary in that adaptation and skill learning appear continuously and on many time scales. The section on motor systems addresses this challenge by the inclusion of complementary approaches, spanning many levels of analysis.

The section begins with a review of motor primitives by Bizzi and Mussa-Ivaldi. This is a critical solution to the degrees-of-freedom problem in limb movement. There are an infinite number of possible muscle activation patterns that could lead to similar movements. How does the nervous system constrain this redundancy? Motor primitives simplify the problem by reducing muscle activations to a set of muscle groups that can be combined, analogous to the mixing of basis functions, to generate characteristic limb movements.

Anatomical studies in the chapter by Dum and Strick extend their previous work demonstrating distinct output channels from different parts of the basal ganglia to distinct cortical areas. New work reveals distinct output channels from the cerebellar nuclei to cortex as well. The existence of distinct cortical projects from basal ganglia and the cerebellum has many important functional implications and could explain the diverse semiology of deficits in patients with lesions to either the basal ganglia or the cerebellum. Alternative models of basal ganglia function, based on both clinical and physiological evidence, are reviewed in greater detail in the chapter by Graybiel and Mink. They present recent evidence for dopamine mediated learning mechanisms, action selection processes and on-line control within the basal ganglia. Particular emphasis is placed on the selection of action based on prior experience combined with contextual information that might influence choice.

Shadmehr and Krakauer consider the problem of action dynamics more broadly. They consider patients with lesions of the cerebellum, parietal cortex, and basal ganglia and interpret their deficits in terms of computational processes such as state estimation, optimization, prediction, cost, and reward. From this evidence, they argue for relative specialization by which the cerebellum builds internal models that predict sensory outcome of motor commands and correct motor commands, the parietal cortex is used for state estimation, and the basal ganglia are needed for learning costs and rewards associated with sensory states.

Evidence for a unique role of the parietal cortex in state estimation, that is, the integration of vision and somatosensory information with motor command is considered in the chapter by Mulliken and Andersen. They emphasize the role of the parietal cortex for generating a forward model that predicts the sensory consequences of a movement. This prediction is likely to be used in a number of action-relevant processes, including on-line control, evaluating performance with a desired outcome, canceling reafferent input, and mental simulation and determining agency.

The computational methods used in state estimation merge sensory feedback and ongoing motor commands into a common theoretical framework. These different sources of information are traditionally considered to represent solutions to very different computational problems. Perception has to do with inferring the state of the world given sensory data and action with generating motor commands that lead to a task goal. In his chapter, Todorov summarizes how these two problems are in many ways related. Sensory inference and motor prediction can be united in a single computational framework. With this, it is possible to explore computational similarities and differences between the two systems.

The overlap between perception and action at the neural level is considered in the chapter by Rizzolatti, Fogassi, and Gallese. Using the mirror neuron as a core mechanism for representing an action that is either executed or perceived, they present evidence that extends this general mechanism to humans, where it could be used to understand intentions in others. Impairment of this process might explain some of the clinical deficits that are seen in autism spectrum disorders.

In the final chapter in this section, Grafton, Aziz-Zadeh, and Ivry consider the importance of hierarchical representation as an organizing principle for understanding how people are capable of creating as well as recognizing the meaning of complex, goal-oriented action. They review functional and behavioral evidence in human experiments of hand-object interactions, bimanual control, and the integration of semantics into action planning. The results support the existence of a highly flexible control hierarchy rather than a strict anatomical hierarchy for organizing complex motor behavior.