3.5 Generation of the fingers trajectories and motion controller
It is worth noticing that when the sum of each contribution in (3.2) for a finger
results in a zero force field, the corresponding contact point does not change its
position in the actual step of the current iteration of the planner stage.
3.5
Generation of the fingers trajectories and mo-
tion controller
The local grasp planner produces a sequence of intermediate target grasp configu-
rations at each iteration of the object's reconstruction algorithm, which ends with
the optimal grasp configuration (in a local sense). The intermediate configurations
are used to generate the paths for the fingertips.
Namely, the sequence of intermediate configurations is suitably filtered by a
spatial low-pass filter in order to achieve a smooth path for the fingers on the object
surface. Notice that only the final configuration needs to be reached exactly, while
the intermediate configurations can be considered as via points for the generation
of the trajectories of the fingers, and that can be computed in real-time with a
one step delay.
With respect to the smooth paths through the points of the filtered configura-
tions, the actual paths of the fingers generated by the trajectory planner keep a dis-
tance
f
along the normal to the surface. When the final configuration is reached,
the safety distance is progressively reduced to zero, producing the desired grasp
action, with directions of the grasp perpendicular to the object's reconstructed
surface.
A kinematic control is used to allow a correct tracking of the trajectories given
by the planner. In particular, a closed-loop inverse-kinematic algorithm [89] has
been exploited to reach such a goal. This method requires as input the desired
position for each finger, and as output it gives the relative joints velocity. High gain
low-level controllers are necessary to physically accomplish the task. Moreover, an
interaction control (e.g. an impedance control) can be used to touch the object
with a desired dynamic, and a force optimization algorithm [12] could be used for
a proper distribution of the grasp forces.
3.6
Simulations and experiments
The proposed method about visual grasp has been experimentally tested on dif-
ferent real objects considering a different number of fingers of the available robotic
hand of Figure 3.7. Obviously, since it is not an anthropomorphic hand, a human-
like grasp here means that it must by stable and the hand should be in a feasible
and dexterous configuration. In the following, the results for the objects shown in
Figure 3.5, namely a teddy-bear and a little bottle, are presented.
37