This fits within the experimental work because coactivation is seen as one of the first responses to changing dynamics whether or not such coactivation is required for the final

adaptation to the dynamics (Franklin et al., 2003, Osu et al., 2002 and Thoroughman and Shadmehr, 1999). Only a limited amount of work has been done so far to investigate the neural underpinnings of impedance control. It has been suggested that the cerebellum is the brain area most likely involved in impedance control (Smith, 1981). This has been supported by changes in cerebellar firing during coactivation (Frysinger et al., 1984) and several fMRI studies investigating the coactivation involved in stabilizing an unstable object compared to a matched

stable object (Milner et al., 2006 and Milner et al., 2007). However, in these see more two fMRI studies, it is not clear that a forward model could be separated from an impedance controller (because both could have been used for the unstable task, but not for the stable task). Earlier work also proposed that there are separate cortical areas for the control of movement/force and joint stiffness (Humphrey and Reed, 1983), a finding supported by psychophysical studies (Feldman, 1980 and Osu PD173074 ic50 et al., 2003), but not conclusively. In terms of the adaptive control of feedback gains that change with the environmental compliance, the results are much clearer. Recent studies using single-cell recordings in monkeys and TMS in humans have shown that these task-dependent feedback gains are dependent on primary motor cortex (Kimura et al., 2006, Pruszynski et al., 2011 and Shemmell et al., 2009). Finally, we examine the issue of learning. As already discussed, one of the features that makes control difficult is nonstationarity. Both over the long timescale of development and aging as well as on the short timescales of fatigue and interactions with objects, the properties of the neuromuscular system change. Such changes require us to adapt our control

strategies—in CYTH4 other words, learn. In sensorimotor control, two main classes of learning have been proposed: supervised learning, in which the (possible vector) error between some target of the action and the action itself drives learning (Jordan and Rumelhart, 1992 and Kawato et al., 1987); and reinforcement learning, in which a scalar reward signal drives learning (Dayan and Balleine, 2002 and Schultz and Dickinson, 2000). The third main type of learning, unsupervised learning, has been a focus primarily in the modeling of sensory processing (Lewicki, 2002 and Olshausen and Field, 1996). There has been extensive work in sensorimotor control suggesting that an internal model of the external environment is learned (for a review see Kawato, 1999 and Wolpert and Kawato, 1998). This has focused on the adaptation of limb movements to novel dynamics.