Model-based studies of control strategies for noisy, redundant musculoskeletal systems

UnrestrictedPrecise control of posture is necessary for the success of any motor actions. The strategies hereby precise movement is achieved reflect many complexly interacting factors, including the tendency of active muscles to generate substantial motor noise as well as to influence the mechanical impedance of a limb with multiple degrees of freedom. A realistic model of the human sensorimotor system would be instrumental in understanding the underlying principles whereby the CNS controls the kinematic effects of signal dependent noise (SDN) in the actuators of a redundant musculoskeletal system. In this work we first constructed a comprehensive sensorimotor system model of the human arm that accurately represents the physiological properties of human muscles and proprioceptors. A simplified two-joint six-muscle arm model was then used for the simulation study of hand postural control in the face of SDN. In single-joint systems, co-contraction has been shown to be effective to reduce the kinematic effect of SDN. But the strategy is energetically inefficient, and extension to a multi-joint system is not obvious because of the intersegmental dynamics and redundant musculature.; In this study, we used Monte Carlo simulation to explore how various patterns of muscle activation affect both the magnitude and direction of such perturbations and the energy consumption under various limb positions and loading conditions. Multiobjective optimization analyses identified a set of control patterns that satisfied both precision and energy criteria. We found that lower postural variability could generally be achieved with higher energy consumption, but the relative levels of activation of each muscle changed drastically at each control level rather than simply scaling proportionately. The end-point variability could be measured and minimized for a task-specific axis, which also affected which strategy provided the best performance, which depended also on limb configuration and external loads. Interestingly, no correlation between end-point stiffness and kinematic variability was found. We conclude that in the presence of SDN, the CNS can utilize muscle redundancy to maximize task performance but that the set of optimal strategies does not follow simple rules such as stiffness modulation and co-contraction, and probably needs be learned rather than computed by the CNS