Dialogue on Perception and Action
Also , it added little to be able to see the ball for the last 300 ms, which would indicate that much of the information for controlling the catch was picked up before that time. C: Sharp ' s and Whiting's results lend support to the suggestion that human arm movements are regulated by a pulse frequency of about 3 Hz. This frequency also corresponds to the physiological tremor at the elbow of human adults (Fox & Randall , 1970). However. the pulsing seems to be a property of the control structure itself rather than a function of some physical constraints. Trevarthen and his colleagues ( 1981) as well as I myself have found the same pulsing of arm movements in neonates. There is further evidence showing that eye movements of adults ( Yarbus , 196 7) are regulated by the same base rhythm. D: It's perhaps better to talk about regulation at rather than by the same base rhythm, don't you think·) .. By .. implies a central clock controlling it all, whereas it may be, for example. that the observed pulsing is due to lag in the perceptuomotor cycle or inertial propet1ies of the biological system. Of course, that raises the question why eye and arm movements should show similar pulsing. C: Eye movements and ann movements seem to have an important property in common. They are both controlled by peripheral vision. When people reach for targets they always fixate the target, never the hand. This is also true for young infants. An experiment reported by Paillard ( 1980) gives further evidence for this position. Subjects who were only allowed foveal vision did not adapt thei r pointing behavior to a prismatic displacement. If instead they were allowed only peripheral vision, they did adapt, but only if they could see the continuous movements of their arms. Paillard argued from these and similar results that there are two sensorimotor systems regulating arm movements. The system I already mentioned controls the direction of the trajectory of the moving hand relative to the stabilized orientation of the visual axis and relies essentially on movement information provided by peripheral vision . The other system controls the accurate homing-in of the hand onto the target and relies essentially on positional information provided by foveal vision. Paillard · s two systems correspond to an important functional distinction made by Bernstein ( 1947) between body-related and object-related movements. There are some beautiful experiments by Buyakas et al. ( 1980) illustrating these points. They placed the subject in a completely dark room and asked him to move his index finger onto a point of light while they monitored his movements. They found that the subject would start out with a rather precise movement which ended. on the average, two degrees of visual angle from the target. However. not being able to see his hand, the subject was lost from there on. Corrective movements were just as often directed away from the target as toward it. The phenomenon has. in fact , been known for some time. As far as I know it was first described by Sandstrom ( 1951). He called it "the eluding light point. , . D: These two control systems you talk about , do you think they function independently of one another'1
11. DIALOGUE ON PERCEPTION AND ACTION 235
C: No , even if it is possible to separate them under special conditions it is obvious that they function as an integrated whole in everyday life. The anticipatory adjustments of the hand are delicately timed relative to the movement of the arm . However, it is not only in reaching and batting that timing is an important aspect. Doesn ' t it seem to enter into most kinds of skilled activities? The timing of a skilled person is delightful to watch. Take for instance downhill skiing. D: Yes .. . did I tell you we 'd recently done a film analysis of ski-jumping (Lee et al. , 1982)? The films show the ski jumper right near the lip of the jump. We were interested in the timing of the explosive straightening of the legs just before the lip because, as you know, it's critically important that they get that right. What we found was the standard error of the timing across 14 jumpers was only 10 ms. C: Do you think the same sort of thing goes on when you're running and jumping-say as in the long jump--that long jump study you did? D: Yes , I do. As you know we did that study a few years ago and the results clearly showed that the jumpers were visually regulating their gait over about the last four strides in order to hit the take-off board . But it puzzled me for a long time as to how they were doing that , what exactly they were doing and what type of visual information they were using. And it wasn ' t really until I started to think about the mechanics of running that it dawned on me that perhaps timing was the central essence of it. You see there was something else to explain too and that was the rather consistent stride pattern the athletes achieved during the initial phase of the run-up. If the strides had been a regular length that would have been fairly easy to understand but since they were accelerating their strides were gradually increasing in length. And then there was the smooth transition from this stereotyped phase of the run-up to the visually regulated phase which made me think that perhaps throughout the whole run-up they were regulating just one parameter of their gait. Anyway to cut a long-jump story short, the conclusion that I came to was that the gait parameter they were regulating was the vertical thrust - or more precisely, the vertical impulse that they applied to the ground at each step. During the stereotyped phase of the run they tried to keep the impulse constant, then , as they neared the board , they regulated the impulses and thereby the durations of their strides on the basis of visual information about the time-toreach the board (Lee et al., 1982). C: There must also be a lot of fine-grained timing. don ' t you think so? D: Certainly. For example , the activities of the leg muscles have to be finely timed relative to the moment of impact of the foot with the ground in order that the leg can act as a shock absorber, as Mel viii Jones and Watt ( 1971) have pointed out. If, for instance, you step off a curb you haven't noticed you can get a very nasty jolt. This type of timing control would certainly seem to be visual. C: There is no doubt that the information for most timing must be visual. Just imagine blindfolding a downhill skier or a tennis player. . . A much more tricky
236 VON HOFSTEN AND LEE
question is to define the information used. You have discussed one potentially useful variable in the optical flow for a subject who wants to time his behavior relative to an object he is approaching or that is approac hing him. D: Yes . in these cases time-to-contact is specified in the optic flow field by the inverse of the rate of dilation of the image of the object (Lee. 1976). People-- and animals too-seem so good at detecting time-to-contact and do it so rapidly that I think it's very likely that they pick up the information directly from the image dilation. C: But time-to-contact is also specified indirectly through the distance to the target and the velocity of approach. Do you have any proof that the system docs not in fact extract information about distance and velocity and work out time from that? D: There are several ways to approach that question. First. there ·s the method Schiff and Detwiler ( 1979) used , showing people movies of an object approaching . No information about the distance or velocity of the object is displayed. only information about time-to-contact through the rate of dilation of the image on the screen. They found that subjects could detect the time-to-contact quite well. though they tended to underestimate. However. the times used - from 2 to I 0 sec-were outside the normal useful range of a second or less required for timing actions like hitting. catching and jumping. Todd ( 1981) used shorter times-tocontact in his computer displays and found that subjects could discriminate (at the 80% level) times-to-contact differing by 50 ms. The experiment testing subjects' ability to actually detect times-to-contact of less than I sec from dis-- plays needs to be done. But it certainly looks as though information about distance and velocity is not necessary for the perception of time-to-contact.