When objects approach on a collision course, young babies will blink their eyes. The timing of the blink is crucial, since it serves to protect the eyes from being injured. A looming virtual object approached infants under different constant velocities and constant accelerations. The youngest infants blinked when the virtual object reached a threshold visual angle, while older infants geared their blinks to the virtual object's time-to-collision. The findings indicate that infants shift to a more sophisticated time-strategy in determining when to make a defensive blink, implying an important neuronal change in the visual system around 6 months of age.

The environment of the newborn abounds with surfaces and objects that may come into contact with the infant's face and cause possible damage to the sensitive cornea of the eyes. Therefore, from very early on in life infants have to be able to sense when something is approaching on a collision course so as to make an appropriate defensive blink. Blinking to visual stimuli is reliably found in infants from about three months of age, and is generally considered the best indicator of awareness to stimuli on a collision course in early infancy (1).

Most experiments on defensive blinking have used a so-called looming stimulus to elicit a response (2), showing clear avoidance responses to expanding virtual visual objects in both animals and humans (3 - 5). Recently, the localisation of looming sensitive neurons in the visual pathways of non-human species has received special attention (6, 7).

Young infants are in need of a strategy that will help them to time their defensive blink perfectly. If they blink too early, they risk having reopened their eyes even before the collision has taken place. If they blink too late, the object would have hit them, possibly injuring their eyes. Are human infants capable of timing their blinks accurately, or are their blinks to be seen as simply reflexive, displaying no evidence of advanced control? To test which timing strategy babies between 5 and 7 months use when blinking to protect their eyes, we measured defensive blinking while a virtual object approached the babies on a direct collision course (8). A colourful, circular, expanding virtual object was projected onto a large sheet and rotated towards the baby under different constant velocities and constant accelerations (Fig. 1). To test which strategy infants use in determining when to make the defensive blink (Fig. 2), we divided the babies into two groups. Group 1 consisted of 5-to-6-month-old infants who all geared their blinks towards a certain threshold value of the virtual object's visual angle (Fig. 3A). This strategy turned out to be successful when the object was approaching at a slow rate. However, when the object approached fast and/or under acceleration, the threshold often lay too close to impact and, as a result, the defensive response frequently occurred too late (Fig. 3B). The slightly older 6-to-7-month-old babies of group 2, on the other hand, geared their defensive blinks to a specific time before the impending collision, irrespective of the virtual object's visual angle and its approach velocity. The older infants thus all blinked at smaller visual angles as the virtual object's approach velocity increased (Fig. 3C), while keeping the time-to-collision constant (Fig. 3D). There was no evidence to suggest that any of the babies blinked at a certain threshold value of angular velocity.

Thus, as infants get older and more experienced, they shift from a less advanced timing strategy based on visual angle to a more advanced strategy based on time-to-collision in determining when to make a defensive blink. These results are in agreement with earlier findings on timing behaviour in animals (9) and humans (10 - 14) and counter the view that the blinking response in young infants is crude and unsophisticated.

Instead, the switch in strategy is likely to be the result of an increasing ability on the part of the infant to pick up relevant information from the environment. Infants show an increased sensitivity to different kinds of energy change in the optic array with age (15). They become better at searching out the relevant information, and the search becomes more systematic and efficient. Information in the optical flow field is always present, but it may take some experience with movements and objects to discover the need to use a more sophisticated strategy. The observed switch in strategy in the present experiment coincides with the development of stereoacuity which implies that some radical change in the neuronal substrate is occurring, a change which is very likely cortical (16). The ability to make precise timing decisions with respect to potentially harmful objects in the environment is especially important as the infant's mobility increases around 6 months of age. As infants become more mobile, they will experience a greater need for making correctly timed responses, which may prompt a switch to a more sophisticated strategy based on time-to-collision.

Research from the field of developmental psychology can often be accused of demonstrating in a rather superficial manner that infants behave more or less like adults (17). It is, however, more realistic to assume that smart perceptual devices must be developed. In such a view, lower level variables might be used initially in timing behaviour, and might guide the search for, or become an integrated part of, higher order variables that take over. Our results suggest that lower level variables, such as visual angle, are used early in development, only to be substituted as the infant's ability and needs change by a more advanced timing strategy based on the higher level variable time-to-collision.

Our findings could have practical implications for the early diagnosis of babies who are neurologically at risk of brain damage because of prematurity or low birth-weight. Research from our laboratory has shown that preterm infants who continued to use, beyond the first year of life, a less advanced timing strategy when catching moving toys, were more likely to be officially diagnosed as suffering from cerebral palsy at age two (18). Since defensive blinking is such an early-developing, fundamental response in infancy requiring precise timing skills, it is very likely to provide detailed information about the neurological status of children born at risk of developing brain damage. Information of this kind at an early stage of the child's development might result in early intervention while the child's brain is still in a plastic state.

References and Notes

1. A. Yonas, in Development of Perception: The Visual System, R. N. Aslin, J. R. Alberts, M. R. Peterson, Eds. (Academic Press, New York, 1981), pp. 313-334.
2. W. Schiff, J. A. Caviness, J. J. Gibson, Science 136, 982 (1962).
3. W. Schiff, Psychol. Monogr. 79, 1 (1965).
4. T. G. R. Bower, J. M. Broughton, M. K. Moore, Percept. Psychophys. 9, 193 (1970).
5. J. E. Náñez, Inf. Behav. & Dev. 11, 447 (1988).
6. Y. Wang, B. J. Frost, Nature 356, 236 (1992).
7. H. Sun, B. J. Frost, Nature Neurosci. 1, 296 (1998).
8. The experiment was based on the fact that the infants believed the portrayed illusion, and thought that something actually was coming towards them. Nine healthy, full-term infants between 5 and 7 months of age took part in the experiment. The infants were sitting in a baby chair with their faces 40cm away from a 2 x 2m plain white sheet on which the looming stimulus was projected. The stimulus consisted of a black circle with four small, brightly coloured circles rotating within it. The virtual object and its looming approach were programmed on a Commodore Amiga 600, and projected onto the sheet by a video projector. The object initially had a diameter of 4 cm (visual angle (θ) = 6 degrees) and grew in 2, 4, or 6 seconds to a diameter of 130 cm (θ = 117 degrees). A small video camera was placed onto the leg of the baby chair, facing the infants so as to get close ups of their faces. Another camera was placed beside the video projector so as to record the virtual object's approach. The two cameras were recorded simultaneously (split-screen), and a stopwatch was added. The virtual object approached the infants under five different, randomly presented conditions, three constant velocity approaches (C2, C4, and C6) and two constant accelerative approaches (A2 and A6). So as to be included in the analysis, the infants had to blink on a minimum of three trials per condition, resulting in at least 15 trials for each infant. For every blink included in the analysis corresponding values for visual angle, angular velocity, and time-to-collision were computed. The virtual object had the same size/visual angle at the beginning and the end of the approach, independent of the object's approach time. The virtual object's approach speed therefore increased as the approach time decreased.
9. D. N. Lee, P. E. Reddish, Nature 293, 293 (1981).
10. W. Schiff, M. L. Detwiler, Perception 8, 647 (1979).
11. D. N. Lee et al., Quarterly J. Exp. Psychol. 35A, 333 (1983).
12. G. J. P. Savelsbergh, H. T. A. Whiting, R. J. Bootsma, J. Exp. Psychol. Hum. Percept. Perform. 17, 315 (1991).
13. A. L. H. van der Meer, F. R. van der Weel, D. N. Lee, Perception 23, 387 (1994).
14. S. K. Rushton, J. P. Wann, Nature Neurosci. 2, 186 (1999).
15. E. J. Gibson, Annual Rev. Psychol, 39, 1 (1988).
16. R. Held, in Neonate Cognition: Beyond the Blooming Buzzing Confusion, J. M. Mehler, R. Fox, Eds. (Lawrence Erlbaum, Hillsdale, NJ, 1985), pp. 37-44.
17. C. Michaels, P. J. Beek, Ecological Psychol. 7, 259 (1995).
18. A. L. H. van der Meer et al., Dev. Med. Ch. Neurol. 37, 145 (1995).
19. We are grateful to all the babies and parents for their good-natured participation. Thanks are also due to Kolbjørn Barmen for technical help and for programming the looming stimulus, and to Kyrre Svarva for help with the statistical analyses.

Figure captions Fig. 1. Drawings of a five-month-old baby taking part in the experiment.

Fig. 2. A theoretical graph of the growth of the visual angle for the three different constant (C2, C4, and C6) and two different accelerative (A2 and A6) approaches of the virtual object over time, showing (horizontal line) how the available time becomes shorter as the approach time decreases (i.e. as the approach speed increases) when the infant uses a particular visual angle to trigger the defensive blink, and (vertical line) how the visual angle becomes smaller as the approach time decreases when the infant uses a particular time to trigger the defensive blink.

Fig. 3. Average angle and time-to-collision values across the virtual object's different approach times for all the infants. The five infants using an angle-strategy (Mean age: 167 days), on both constant velocity (C) and constant accelerative approaches (A), blinked (A) roughly at the same size of the visual angle irrespective of the virtual object's approach time, while they (B) varied the time-to-collision over the different approach times. Note that the infants using an angle-strategy on average responded too late on the fastest accelerative trials (A2). As a result, no meaningful corresponding visual angles could be computed for these late trials. The four infants using a time-strategy (Mean age: 188 days), on the other hand, (C) varied the timing of their blinks to larger visual angles as the virtual object's approach time increased from 2 to 4 to 6 seconds, while they (D) blinked at roughly the same time-to-collision. The infants using a time-strategy successfully coped with all the virtual object's approach conditions, whereas the significantly younger infants using an angle-strategy did not, t(3.957)=2.54, p < 0.04. Two within subjects analyses of variance (ANOVA) showed significant group x approach time interaction effects for both the constant velocity approaches, F(2,6) = 6.86, p < 0.03, and the constant accelerative approaches, F(1,3) = 59.65, p < 0.005, indicating that the infants using an angle strategy blinked at larger values of time-to-collision as the virtual object's approach time increased (i.e. as the approach velocity decreased), whereas the infants using a time-strategy kept the timing of the blink relatively constant over the different approach conditions.