Neuroscience Letters 334 (2002) 13–16
www.elsevier.com/locate/neulet
The visual control of ball interception during human locomotion
A. Chardenon a, G. Montagne a,*, M.J. Buekers b, M. Laurent a
a
Faculty of Sport Science, Movement and Perception Laboratory, Marseille, France
Faculty of Physical Education and Physiotherapy, Motor Learning Laboratory, Leuven, Belgium
b
Received 17 May 2002; received in revised form 2 September 2002; accepted 2 September 2002
Abstract
According to the required velocity model, on-line modulations of movement acceleration are performed on the basis
of an optically specified difference between required and current behavior. Can this model account for observed displacement regulations in an interceptive task requiring locomotive displacements? In the present study, a virtual reality setup was coupled to a treadmill. Subjects walking on the treadmill were required to intercept a virtual ball approaching at
eye-level by adjusting their velocity, if necessary. While the required velocity model could partially account for displacement regulation late in the interception, it was ineffective to explain early regulations. The possible use of a bearing angle
strategy to control displacement regulation and the possible degree of complimentarity of these strategies are
discussed.
q 2002 Elsevier Science Ireland Ltd. All rights reserved.
Keywords: Perception-action coupling; Locomotion; Law of control; Bearing angle strategy; Goal-directed action
Intercepting or avoiding a moving object is one of the
most complex abilities characterizing the human perceptivo-motor repertoire. It requires on-line movement regulations, based on instantaneously available information, to
accommodate to a future event [9]. This is the case when
driving a car, when moving through a crowd, or when intercepting a moving ball. While the information substrate
possibly used in this kind of task has received special attention [10,14], very few integrative models exist which
describe not only the information but also how the information is used in the control process.
The required velocity model [1,2,13] constitutes a rather
isolated attempt to bridge the gap between perception and
action. According to this model, on-line modulations of
movement acceleration are performed on the basis of an
optically specified difference between required and current
behavior (Eq. 1). This continuous updating of the movement
with respect to current information [15] guarantees success.
In one study [13] a one-handed catching task was used in
which the hand was constrained to move along a single
lateral direction while the angle of approach of the ball
* Corresponding author. Université de la Méditerranée, Faculté
des Sciences du Sport, UMR Mouvement et Perception, 163
Avenue de Luminy CP 910, 13288 Marseille Cédex 9, France.
Tel.: 133-49-1172273; fax: 133-49-1172252
E-mail address: [email protected] (G. Montagne).
was varied. In this model, the required behavior corresponds
to a required hand velocity expressed by the ratio between
the current lateral distance separating the hand from the ball
and the time remaining before the ball reaches the movement axis (Eq. 2) (see [5] for an optical specification of this
ratio). The current behavior corresponds to the actual velocity of the hand.
Y
Y
X YY 2 aXreq
2 bXcur
Y
Xreq
¼
Xb 2 Xh
TC
ð1Þ
ð2Þ
Y
where X YY is hand acceleration, Xreq
is the required velocity,
Y
Xcur is the current velocity, Xb is the current lateral ball
position, Xh is the current hand position, TC is the time
remaining before the ball reaches the movement axis and
a and b are constants.
According to this model, all the subject has to do is to
adjust hand acceleration to cancel the difference between
required and current behavior.
Recently this model received empirical validation
[11,12]. In these two experiments the subjects were asked
to intercept moving objects while various parameters were
manipulated, e.g. angle of approach, ball speed, duration of
presentation. The results show adaptations of hand kinematics that match the predictions made on the basis of the
0304-3940/02/$ - see front matter q 2002 Elsevier Science Ireland Ltd. All rights reserved.
PII: S03 04 - 394 0( 0 2) 01 00 0- 5
14
A. Chardenon et al. / Neuroscience Letters 334 (2002) 13–16
Fig. 1. Bird’s eye view of the interceptive task. The subject intercepts a moving ball with the head by producing the necessary
displacement regulations. The required velocity takes into
account the distance separating the current position of the
subject from the current projection of the ball on the axis of
displacement (Yb 2 Ys) and the current time remaining before
the ball reaches the axis (TC). The bearing angle b is also represented.
model. For example, in a ball-catching experiment [12], it
was shown that the cancellation of the difference between
required and current behavior occurred on average 300 ms
before contact and was maintained until contact. Moreover,
if the subject’s hand was initially placed in the correct position to catch the ball (i.e. no movement necessary), movements away from this position occurred as soon as a lateral
hand–ball distance was created, followed by reversal movements back to the correct position.
These results argue for the validity of the model under
consideration, nevertheless they call for a few remarks. First
of all, they were obtained using a task in which hand movements were mechanically constrained to one direction.
While this added constraint can be justified insofar as it
allows a more easy test of the model, it remains to be
seen if the model applies to other, more realistic tasks. It
is also worth mentioning that in the three previously cited
studies [11,12,13], the movement times were between 1 s
and 1.5 s. Consequently, one can ask if the same model can
account for movement regulation when the interceptive task
involves a larger time scale (around 8 s in the present experiment).
The present study was designed to explore the utility of
the required velocity model for a longer time frame with a
different task. Subjects were asked to move (locomotion
displacement) in a specified direction (along the Y-axis)
and to intercept with their head a virtual moving ball crossing their displacement axis. The required velocity model can
be simply reformulated by expressing both required and
produced behavior along the Y direction (Fig. 1).
Eight subjects with normal or corrected to normal vision
participated in the present study. Informed consent was
obtained prior to testing. The experimental set-up consisted
of a virtual environment (Silicon Graphics) coupled to a
treadmill (Gymroll). Subjects walked on the treadmill 0.6
m in front of the screen (3 £ 2.3 m) on which the virtual
scene was projected. The virtual scene consisted of an
approaching ball (0.2 m diameter) and a textured floor.
Displacements of the virtual environment were directly
proportional to the displacements of the subject on the treadmill. Ball path kinematics were collected by the Silicon
Graphics system (100 Hz) and subjects’ displacement kinematics were registered by an optical encoder positioned on
the treadmill (100 Hz). The synchronization of the two
systems allowed the computation of all displacements
(ball and subject) in the same coordinate system. This setup was chosen for obvious reasons: the task constraints are
easy to manipulate while natural movement regulation
behavior is maintained [4]. Subjects were required to intercept the ball approaching along a rectilinear path (26.568
with respect to subjects’ displacement axis) with a constant
velocity (1.26 m s 21) along a horizontal plane going through
the head. The interception point was computed at the beginning of each trial on the basis of subjects’ initial velocity
(mean velocity 0.81 m s 21, SD 0.057 m s 21), while the time
taken by the ball to cross subjects’ displacement axis was
kept invariant (7.5 s). Subjects performed the experimental
task in three conditions: no change, acceleration and deceleration. In the no change condition, no velocity adjustments
were required. In the other two conditions subjects had to
either accelerate, or decelerate to intercept the ball. More
precisely, in the acceleration and deceleration conditions, if
the subject kept his velocity unchanged he would have been
one meter short or ahead of the interception point, respectively. Subjects performed ten trials per condition, the order
of conditions was randomized across all trials. Subjects
received immediate performance feedback after each trial:
a green (success) or red (failure) square projected in the
center of the screen. A successful trial was one where the
distance between the center of the ball and the center of the
subject’s head was at most ^2 cm when the ball crossed the
subject’s displacement axis.
Current error was computed every 10 ms during each trial
and represents the error that would result if the current
velocity had been kept invariant. Consequently, a current
error that tends towards zero implies effective regulation of
velocity. Performance measures were percentage of
successful trials and the time-course of inter-trial variability
of both velocity and current error (i.e. each half second
during the last 6 s before contact). An increase in velocity
variability indicates velocity regulations. In the assumption
that these velocity regulations are effective, they should be
accompanied by a decrease in current error variability. For
the required velocity model analysis, the fit of the model to
the recorded data was examined using multiple linear
regression on each trial. The analyses were based on Eq. 1
and tested the extent to which acceleration depends upon the
A. Chardenon et al. / Neuroscience Letters 334 (2002) 13–16
Fig. 2. Time course of both mean displacement velocity and
current error profiles with the associated variability (error bars)
in the acceleration condition (diamond), the no change condition
(square) and the deceleration condition (triangle). The analyses
reveal an initial increase in velocity variability 5 s before contact
and a second increase 1.5 s before contact together with a
decrease in the current error variability from 5 s before contact.
Fig. 3. Mean bearing angle profiles for every successive 0.5 s
interval from 6 s until 1 s before contact, and their associated
variability. The analyses reveal that the bearing angle is kept
constant in the first part of the trial. Bearing angle only changes
in the last 2 s before contact.
15
velocity differential. The values of the two parameters of
Eq. 1 (i.e. a and b) were not set a priori but corresponded to
the values that led to the better fit. The validity of the model
was assessed by looking at the average percentage of the
total variance explained by the model.
An initial ANOVA showed that the percentage of
successful trials was not affected by experimental condition
(Fð2;14Þ ¼ 1:125, P . 0:05). Subjects’ success rate was 61%
in all three conditions (acceleration, deceleration and no
change). This relatively poor success rate should not be
too surprising as it is the result of the particularly stringent
success criterion characterizing the task (spatial window: 4
cm, temporal window: around 30 ms). Subsequent
ANOVAs were conducted on all trials (i.e. successful trials
and failures, n ¼ 240). Velocity and current error variability
profiles indicated an initial increase in the velocity variability 5 s before contact and a second increase 1.5 s before
contact (Fð11;77Þ ¼ 34:085, P , 0:001) together with a
decrease in the current error variability 5 s before contact
onward (Fð11;77Þ ¼ 53:646, P , 0:001) (Fig. 2). These
results reveal effective velocity regulations occurring very
early in each trial.
The required velocity model analysis was conducted over
the 240 trials. Multiple linear regressions were performed
over six temporal ranges: 6 s before contact onward, 5 s
before contact onward,..., 1 s before contact. These analyses
showed that the percentage of total variance (R 2 £ 100)
explained by the model during the last 6 s of the task was
an inverse function of the remaining time (8.5%, 9.5%,
11%, 14.5%, 25% and 46% for the time periods 6, 5, 4,...,
1 s before contact, respectively). An analysis of variance
performed on the data after Z-transformation revealed an
increase of the Z values 3 s before contact onward
(Fð5;35Þ ¼ 261:294, P , 0:001). Hence, the required velocity model only predicts the observed kinematics in the
vicinity of contact, accounting for (at best) 46% of the
variance and cannot account for early velocity regulations
(i.e., at 5 s before contact). This rather weak percentage of
variance explained by the model could also highlight the
need in future modeling work to integrate in some way a
noise term in order to account for the inter trial variability
inherent to the production of such behavior.
An additional analysis was performed in order to see
whether an alternative strategy of maintaining a constant
bearing angle [3] could explain the observed early velocity
regulations. Bearing angle is the angle subtended at the
point of observation by the current position of the ball and
the direction of displacement (Fig. 1). Linear regression
analyses were performed for each trial (n ¼ 30) performed
by a representative subject on the time-course of the bearing
angle from the beginning of the trial until 2 s before contact.
If the bearing angle strategy had been used, the regression
analyses should have zero gradient. The analyses revealed,
for each trial, variations of the bearing angle across time
(P , 0:05 and mean R 2 values 0.73). Nevertheless, the
slopes were not statistically different from zero
16
A. Chardenon et al. / Neuroscience Letters 334 (2002) 13–16
(tð1;29Þ ¼ 22:024, P . 0:05)(mean value 20.0009, SD
0.0025). Thus, the bearing angle strategy can explain the
early velocity regulations. It is worth noting that one can not
exclude that this strategy was also used to produce late
velocity regulations. In fact the large bearing angle variations in the last second before contact (Fig. 3) could express
small spatio-temporal errors that get magnified at small
distances when plotted in term of bearing angle.
The present study was designed to test the applicability of
the required velocity model over a longer timescale than in
previous studies. The percentage of variance explained by
the model is slightly lower than the one obtained in unimanual interceptive tasks [12] (46% vs. 55%). Our data
show that this model has the potential to explain part of
the velocity regulations occurring late in the task. Conversely, the first part of our analysis also reveals the existence
of visually driven velocity regulations early in the task,
which the required velocity model cannot explain. For
these regulations, a simple strategy based on maintaining
a constant bearing angle is adequate. It should be noted
that we do not want to exclude the possible involvement
of this strategy in producing late regulations. However the
present data do not permit us to provide an answer to take a
clear position in this matter.
This study does not support the required velocity model
as a generic model for the control of interceptive tasks.
Conversely it allowed us to identify a simple bearing
angle strategy as a suitable mechanism to perform the ball
interception task (see Refs. [7,8] for a similar result in a
goal-directed cycling task). However, it remains to be
seen if the produced regulations can be explained by the
use of a single control mechanism based on the bearing
angle, or if they necessitate two mechanisms to effectuate
separate control of velocity regulations early and late in the
task. Note that, contrary to the bearing angle strategy, the
required velocity model is based (partly) on the use of
expansion information [5]. Given the task constraints (ball
diameter 0.2 m, combined velocity of the ball and the
subject: 2 m s 21, subject-ball distance: 15 m) the expansion
velocity of the ball at the beginning of the trial (0.05 8 s 21) is
below the perceptual threshold (0.08 8 s 21) [6]. This could
explain why subjects would use a bearing angle strategy
initially and then could opt for a required velocity strategy
when the expansion pattern is easily perceptible, i.e. at the
end of the trial. A simple way to test the two hypotheses
would consist in manipulating the expansion information
during the last seconds before contact. The use of a bearing
angle strategy should not suffer from this manipulation.
The present study illustrates the adaptive capacity of the
perceptual-motor mechanisms underlying goal-directed
locomotion. Rather than emphasizing the (real) limits of
the required velocity model we prefer to keep in mind the
diversity of the mechanisms available. Discovering the
range of action of these mechanisms and eventually how
these mechanisms cooperate offers a very interesting challenge for future work.
[1] Bootsma, R.J., Fayt, V., Zaal, F.T.J.M. and Laurent, M., On
the information-based regulation of movement: things
Wann (1996) may want to consider, J. Exp. Psychol. Hum.
Percept. Perform., 23 (1997) 1282–1289.
[2] Bootsma, R.J., Ecological movement principles and how
much information matters, In A.A. Post, J.R. Pijpers, P.
Bosch and M.S.J. Boschker (Eds.), Models in Human Movement Science, Print Partners Ipskamp, Enschede, 1998, pp.
51–63.
[3] Cutting, J.E., Vishton, P.M. and Braren, P.A., How we avoid
collisions with stationnary and moving obstacles, Psychol.
Rev., 102 (1995) 627–651.
[4] De Rugy, A., Montagne, G., Buekers, M.J. and Laurent, M.,
The study of locomotor pointing in virtual reality: the validation of a test set-up, Behav. Res. Meth. Ins., C, 32 (2000)
215–220.
[5] Laurent, M., Montagne, G. and Durey, A., Binocular invariants in interceptive tasks: a directed perception approach,
Perception, 25 (1996) 1437–1450.
[6] Lee, D.N., A theory of visual control of braking based on
information about time-to-collision, Perception, 5 (1976)
437–459.
[7] Lenoir, M., Musch, E., Janssens, M., Thiery, E. and Uyttenhove, J., Intercepting moving objects during self-motion, J.
Mot. Behav., 31 (1999) 55–67.
[8] Lenoir, M., Savelsbergh, G.J., Musch, E., Thiery, E., Uyttenhove, J. and Janssens, M., Intercepting moving objects
during self-motion: effect of environmental changes, Res.
Q. Exerc. Sport, 70 (1999) 349–360.
[9] McBeath, M.K., Shaffer, D.M. and Kaiser, M.K., How baseball outfielders determine where to run to catch fly balls,
Science, 268 (1995) 569–572.
[10] McLeod, P. and Dienes, Z., Running to catch the ball,
Nature, 362 (1993) 23.
[11] Montagne, G., Laurent, M., Durey, A. and Bootsma, R.J.,
Movement reversals in ball catching, Exp. Brain Res., 129
(1999) 87–92.
[12] Montagne, G., Fraisse, F., Ripoll, H. and Laurent, M.,
Perception-action coupling in interceptive task: first order
temporal relation as an input variable, Hum. Movement
Sci., 19 (2000) 59–72.
[13] Peper, C.E., Bootsma, R.J., Mestre, D.R. and Bakker, F.C.,
Catching balls: how to get the hand to the right place at
the right time, J. Exp. Psychol. Hum. Percept. Perform., 20
(1994) 591–612.
[14] Savelsbergh, G.J.P., Whiting, H.T.A. and Bootsma, R.J.,
‘Grasping’ tau, J. Exp. Psychol. Hum. Percept. Perform.,
31 (1991) 1655–1663.
[15] Warren, W.H., Visually controlled locomotion: 40 years
later, Ecol. Psychol., 10 (1998) 177–219.