Figure 1.

In the experiment, subjects used their muscle activity to perform a goal-directed task with a virtual arm. After a period of familiarization, the coefficient of variation (CV) of subjects’ motor commands (as reflected by muscle activity linear envelopes) was increased by adding signal-dependent noise, and in another condition the CV was decreased with an exponentially-weighted moving average filter. The shading highlights the control loop hypothesized to take precedence in the virtual arm task, with processes associated with human physiology shaded in light blue, and processes in the virtual world shaded in light red. Note that during the experiment the actual human arm does not move due to an external mechanical constraint, and although subjects received proprioceptive feedback about their actual arms, this feedback did not reflect the actions of the virtual muscles and arm, and therefore was not of high relevance to the task (in contrast to the visual feedback).