CHAPTER 2. State of the art
and so on. In addition, both the planner and the controller need to be robust with
respect to the noise of the sensors.
Starting from these considerations, in this survey, a collection of papers dealing
with the control of a single robotic hand has been considered, without any discus-
sion about problems arising with the mechanical construction of robotic hands or
with the kinematics and the constraints in the "hand + object" system. In [8, 67],
an interesting starting point to cope with these issues can be found.
Assuming this background, starting from the early '90s, many approaches for
the control of robotic hands have been reviewed. The major control approaches
noticed throughout the referenced papers can be subdivided in these main classes,
which are namely:
· Computed torque control.
· Event-driven control.
· Hybrid position/force control.
· Hierarchical control.
· Predictive control
· Control basis approach.
· Adaptive control.
· Stiffness control.
· Impedance control.
· Kinematic control.
In the following, a brief description of these classes is reported.
2.3.1
Computed torque control
The computed torque approach relies on the precise knowledge of the model of the
system, which is composed by the robotic hand plus the object and the related
constraints. Canceling the whole nonlinear dynamics of the system, it is possible
to asymptotically track a desired trajectory for the object's center of mass or, with
the assumption of rigid object, for any other point of the object. In the practice,
the knowledge of the model, friction parameters, location of the center of mass,
etc., is not available and this technique can be useful in simulations. Hence, the
need of a precise model for the whole system often involves the use of some other
devices to manage robustness problems like adaptive control schemes.
10