Coordinating movements using optimal control: A neuro-computational perspective

Much of the research on human motor control has examined how humans move a single body part, such as the arm or the eyes. Our everyday actions, however, consist of movements of multiple body parts at the same time. For example, we use both hands to tie our shoes. Even the simple act of raising one arm is accompanied by changes in leg muscles to stabilize stance. For many patients with neurological strokes or disorders, these complex movement skills often cause problems even after simple movements of isolated limbs have recovered. Compared to single limb movements, however, we know very little about how the nervous system achieves coordination between movements of different body parts. We have gained some insights into how the brain coordinates complex movements through experiments in which human volunteers are instructed to produce two different movements at the same time. These experiments have identified fundamental limitations in our ability to move two limbs independently. For example it is quite difficult to rub one's belly and to simultaneously pat one's head. Further experimentation however, has made clear that these limitations are generally not hard-wired into the motor system. Rather they depend crucially on what the goals of the movements are. For example, movements of the two hands can depend strongly upon one another when we manipulate a single object with two hands, and they can be controlled quite independently when we reach out for two separate objects. To understand how the brain coordinates movements in these different situations, we will study how human volunteers learn motor tasks in which the hands must be coordinated to achieve different goals. We will apply approaches from engineering and control theory to predict how, given a specific task goal, an organism should optimally coordinate movements, and will then test these predictions. We will also investigate how specifically the brain achieves optimal control in coordination. For example, what signals are exchanged between regions of the brain that control different movement components? Does coordination depend on 'muscle memory' / a memorized set of activation patterns in particular muscles? Or do different body parts 'talk' to each other by exchanging information about their current position and velocity? How do the ways in which these areas communicate change as coordination skills become more automatic through training? Functional magnetic resonance imaging (fMRI) will be used to measure brain activity in human volunteers while they perform coordination tasks. By changing the goals and requirements of the task, we will be able to identify neural areas that are involved in coordinating movements and specify their function. For example, when we grasp an object we have to coordinate the arm movement and the opening of the hand for grasping. Which neural areas are involved in estimating how far the arm has moved already towards the target, an important signal for deciding when to start the grasp? Which neural areas take these estimates and initiate the opening of the hand? Understanding how the brain coordinates movements and which areas of the brain are involved in this task is essential for understanding the coordination problems faced by patients affected by strokes or brain disorders. This work will provide a theoretical foundation that can lead to new treatments and rehabilitation programs designed to facilitate the ability of these individuals to produce complex motor skills, those that go beyond the control requirements for producing simple movements.

How does the brain coordinate movements? Our approach to this question is to first determine analytically for a given task goal the optimal control policy, a rule by which estimates of a moving limb's state is transformed into motor commands. Volunteers will learn several coordinated movements and we will then identify the control policies they use. For example, using a learned coordination of hand and thumb, we ask which state estimates of the arm govern the movement of the thumb. By using a robotic device to exert forces to the arm, we can alter the muscle commands required to produce a kinematically identical arm movement. If coordination is learned as a memorized pattern of muscle activations, then corresponding changes should occur in the thumb movement. If kinematic or more abstract state estimates are used, the thumb movement should remain invariant. We will also study the control policy used to move a single target with movements of two arms. Using robotic devices, we will perturb one arm and test the prediction that similar corrections should occur in both the perturbed and unperturbed arms. Electromyography (EMG) will be used to determine the timing of these responses, indicating the level of the nervous system at which this control arises. Functional magnetic resonance imaging (fMRI) will be used to determine the neural areas that underlie coordination in these tasks. We will compare a situation in which movement components must be coordinated to a condition in which the same movement components are performed independently within a narrow time window. Kinematic and kinetic measurements will be obtained using an fMRI-compatible robotic arm and force sensor. By using bimanual skills and varying the roles of the two hands, we can determine which areas are involved in state estimation and which are involved in implementing a new control policy.