 
| |
        |
Department of Computer Science ResearchDevelopmental Psychology
ALL collaborates with researchers from development psychology who
study sensorimotor development, problem solving, and representational
capabilities of infants and toddlers. In particular, studying the
development of motor skills of infants learning to perform
goal-directed arm movements provides valuable insight into how very
sophisticated learning systems actually operate. From an artificial
intelligence point of view, we are interested in the policies infants
develop when learning and the methods by which they learn certain
movements. Measurements of gaze and kinematics of reaching provide
means by which the infant's policy and learning mechanisms can be
speculated. Researchers test how the infants reach for an object under
different conditions, such as with and without visual feedback, to see
what changes in reaching kinematics occur. Members of ALL apply
knowledge of reinforcement learning and computational methods to
contribute in theorizing how infants learn certain skills. The
following are brief descriptions of two of the several developmental
psychological research studies we found particularly interesting. More
detailed descriptions of the research being conducted through the
developmental psychology group can be found the developmental
psychology webpage
Infants have demonstrated capabilities to plan arm movements according
to the goal of those movements. A spoon was filled with applesauce
and then presented to infants such that the handle of the spoon was
sometimes oriented to the left, and sometimes oriented to the
right. Younger infants did not seem to take the goal of the action
(getting applesauce to mouth) into consideration when planning their
reach for the spoon. Regardless of the orientation of the handle,
these infants would use the strategy of reaching with their preferred
hand, which half of the time resulted in awkward grasps causing them
to spill most of the applesauce. However, the older infants did take
the orientation of the handle into consideration when planning their
reach for the spoon. The older children would switch hands, depending
on the orientation, such that they always reached with the most
efficient grip (a radial grip where the thumb is directed toward the
bowl of the spoon). Using this grip, they were able to smoothly and
efficiently transport the applesauce into their mouth without spilling
it. The infant learns this process by interacting with the spoon and
remembering what has worked best and what has not worked. This study
suggests that humans learn by trial-and-error and anticipate task
constraints through the progressive incorporation of perceptual
information.
In other studies, researchers investigated whether toddlers could
understand that a partially observable barrier would affect the
environment that the infant could not perceive directly. In these
experiments, a ball rolled down a track; while the initial path of the
ball was observable, the track continued behind a short wall so that
the ball and the track were unobservable. A barrier, which was taller
than the wall and therefore partially observable, was placed on the
track and would stop the ball at a point behind the wall. Where the
ball stopped depended on where the barrier was placed. Under some
conditions toddlers could infer that the barrier did, indeed obstruct
the ball from rolling down the ramp. Surprisingly, under most
conditions, toddlers have a difficult time choosing the correct door
that the ball should be behind, despite the findings of infant looking
studies that which indicate that they should have no difficulty
understanding that the wall should stop the ball. Research continues
to be done to try and understand the factors that might be
contributing to their poor performance at this task.
|