Delays in the transmission of sensory and motor information prevent er
rors from being instantaneously available to the central nervous syste
m (CNS) and can reduce the stability of a closed-loop control strategy
. On the other hand, the use of a pure feedforward control (inverse dy
namics) requires a perfect knowledge of the dynamic behavior of the bo
dy and of manipulated objects. Sensory feedback is essential both to a
ccommodate unexpected errors and events and to compensate for uncertai
nties about the dynamics of the body, Experimental observations concer
ning the control of posture, gaze and limbs have shown that the CNS ce
rtainly uses a combination of closed-loop and open-loop control. Feedf
orward components of movement, such as eye saccades, occur intermitten
tly and present a stereotyped kinematic profile, In visuo-manual track
ing tasks, hand movements exhibit velocity peaks that occur intermitte
ntly. When a delay or a slow dynamics are inserted in the visuo-manual
control loop, intermittent step-and-hold movements appear clearly in
the hand trajectory. In this study, we investigated strategies used by
human subjects involved in the control of a particular dynamic system
. We found strong evidence for substantial nonlinearities in the comma
nds produced. The presence of step-and-hold movements seemed to be the
major source of nonlinearities in the control loop. Furthermore, the
stereotyped ballistic-like kinematics of these rapid and corrective mo
vements suggests that they were produced in an open-loop way by the CN
S. We analyzed the generation of ballistic movements in the light of s
liding control theory assuming that they occurred when a sliding varia
ble exceeded a constant threshold. In this framework, a sliding variab
le is defined as a composite variable (a combination of the instantane
ous tracking error and its temporal derivatives) that fulfills a speci
fic stability criterion. Based on this hypothesis and on the assumptio
n of a constant reaction time, the tracking error and its derivatives
should be correlated at a particular time lag before movement onset. A
peak of correlation was found for a physiologically plausible reactio
n time, corresponding to a stable composite variable. The direction an
d amplitude of the ongoing stereotyped movements seemed also be adjust
ed in order to minimize this variable. These findings suggest that, du
ring visually guided movements, human subjects attempt to minimize suc
h a composite variable and not the instantaneous error. This minimizat
ion seems to be obtained by the execution of stereotyped corrective mo
vements.