We introduce a new EMG state analysis to test two competing hypotheses abou
t the role of muscle coactivity in learning a complex, multijoint reaching
movement. Following Bernstein, one hypothesis is that as a task is learned,
coactivity should decrease as degrees of freedom are released and limb sti
ffness is reduced. An alternative hypothesis is that as movement speed incr
eases with learning, muscle coactivity should increase, possibly to stabili
ze joints against high inertial forces. Three participants performed a vert
ical reaching movement identical to that used by Schneider et al. We monito
red the activity of four arm and shoulder muscles as participants completed
100 practice trials. Each frame of EMG activity was assigned to one of 16
possible combinations of the four monitored muscles based on an on-off acti
vation threshold. This analysis yielded a time-based summary of muscle coac
tivity during the movement and across practice trials. Results of the state
analysis supported the second hypothesis. As participants decreased their
movement times over practice, coactivity increased - participants used more
three- and four-muscle coactivity states. Changes were especially dramatic
during the braking phase of the Up and Down portion of the vertical moveme
nt. When participants performed deliberately slow movements after speeded p
ractice, three- and four-muscle coactivity was suppressed. We suggest that
increased use of muscle coactivity may serve to counteract unwanted rotatio
nal forces generated during fast movements.