CHAPTER 2. State of the art
In [32], two general optimal criteria are introduced, where the total finger
force and the maximum finger force are considered, while in [64] simple geometric
conditions to reach an optimal force closure grasp both in 2D and in 3D are found.
The geometric properties of the grasp are also used in [55] to define some quality
measures, as well as suitable task ellipsoids in the wrench space of the object
have been proposed to evaluate the quality of the grasp also with respect to the
particular manipulation task to accomplish.
A geometrical approach obtaining at least one force closure grasp for 3D dis-
cretized objects is studied in [84], where two algorithms are investigated: the first
finds at least one force closure grasp, while the second optimizes it to get a locally
optimum grasp.
Measures depending on the configuration of the hand [88] define a set of quality
measures based on the evaluation of the capability of the hand to realize the
optimal grasp. A rich survey of these grasp quality measures can be found in [95].
In order to plan a grasp for a particular robotic multi-fingered hand, quality
measures depending both on the geometry of the grasp and on the configuration
of the hand should be taken into account. Few papers address the problem of
grasping an unknown object using a given robotic hand, able to reach the desired
contact points in a dexterous configuration [11, 15, 33, 36, 39].
A grasp control task is considered in [75], where several controllers are combined
to reach different wrench closure configurations, while in [78] grasp prototypes
generalization of example grasps are used as starting points in a search for good
grasps.
2.2.3
Human-like grasp
In general, human beings can grasp and manipulate a large variety of objects with
a high level of dexterity. An elaborated taxonomy of human grasps can be found
in [22].
Throughout the literature, two main approaches among the others can be rec-
ognized in the process of transferring human manipulation skills to robotic multi-
fingered hands. Namely, they are:
· Programming by demonstration.
· Neural networks and genetic algorithms.
Considering the former technique, movements of human beings are recorded
and analyzed off-line using a motion capture system, and therefore the motion
is transferred to a robotic hand [73]. Further, it has been demonstrated that
humans perform different grasps in reason of the task's specifications, even if the
orientation and the location of the objects are kept the same. A programming
by demonstration system, which shows how fine manipulation tasks (e.g. screw
8