Chapter 2. Prehension

80 THE PHASES OF PREHENSION
Again, they could not see their own hand. Subjects were relatively
inaccurate in reproducing orientation. Mean angular errors ranged
from 3 degrees to 5 degrees, and a systematic bias was observed in the
counterclockwise direction. The relative inaccuracy of matching
orientations can be contrasted with the accuracy with which subjects
matched object size.
In summary, replanning for extrinsic properties like object location
takes about 100 ms; this is a much shorter time than the 300 ms
required for replanning grasping based on visual perturbations of
intrinsic object properties like size and shape. Further, visually
matching seen object orientations with the unseen hand is inaccurate
and biased, in contrast with matching seen object diameters.
4.4 Knowledge Of Task Requirements
Over years of practice grasping objects, humans develop a wealth
of knowledge about object function and behavior and this knowledge
can be used to anticipate and predict the results of interactions. In
Chapter 2, we saw that opposition types are a function of both the
object characteristics and the task requirements. In the absence of a
specific tool (e.g., a hammer) one may substitute another object (e.g.,
shoe) with a grasp appropriate for the task at hand, given object
properties. In contrast, with functional fixedness, one is unable to see
other uses of an object designed for a specific function.
Knowledge about the behavior of an object was seen in Figure
2.6. When asked to grasp a mug by its handle, subjects anticipated a
torque acting on the mug caused by the posture not being placed
around the center of mass. Virtual finger 3 was placed in such a way
to counteract this torque. The reason this occurs in the first place is
task-related. While the mug can be grasped in a variety of places,
grasping it by the handle makes the lip of the mug accessible to the
mouth, while protecting the hand from possibly being scalded by a hot
liquid. Thus, planning involves knowledge of task requirements, es-
timation of object properties relative to the task (such as center of
mass), and anticipation of object behavior during the interaction.
As well, humans have implicit knowledge of task mechanics. For
example, to grasp a square block using pad opposition, one finger can
approach the block and push it into the other finger (Mason, 1985).
Finger trajectories could be chosen so that a frictional sliding force
will orient and center the block in the fingers. The alternative would
be to push the block out of the grasp. Knowledge about task
mechanics is advantageous in unstructured unpredictable