CHAPTER 3. Visual grasp algorithm
Image
Acquisition
Object Model
Recontruction
Optimal Global
Grasp
Evaluation
Trajectory
Planning
Grasp
Execution
Grasping
T
ime
Image
Acquisition
Dynamic
Object Model
Recontruction
Real-Time
Grasp and
Trajectory
Planning
Grasp
Execution
Grasping
T
ime
Saved
T
ime
Figure 3.1: Classical serial method vs. proposed parallel approach.
Simulation results and experiments are presented to show the performance of
the visual grasp algorithm. The results presented in this chapter can be also found
in the following papers [56, 57, 58, 59, 61].
3.1
Description of the visual grasp algorithm
A typical approach for the grasp of unknown objects, here called serial approach,
mainly consists of two stages: first, the object is completely reconstructed, and
then the evaluation of the optimal grasp starts under a selected global criterion,
as showed in the Figure 3.1. This approach gives the best results in terms of the
quality of the grasp, since the evaluation is made in a global way. However, the
total execution time is given by the sum of the time due to the reconstruction of
the object and the time due to the synthesis and planning of the grasp.
Although modern technologies allow a fast reconstruction of the object, the
investigation of all the possible combinations of points for the grasp or of the set
of surfaces which approximate the object (depending on the reconstruction method
adopted) could generally require a considerable amount of time. Obviously, this
drawback is irrelevant for off-line applications, but it could be unsuitable for real-
time grasping, if powerful hardware is unavailable.
The proposed method, here referred as parallel approach, may represent a valid
alternative in such cases, where the total computational time is given by the slower
between the reconstruction and the planning stage (see Figure 3.1). As a matter of
fact, these two parallel processes are independent and they can be allocated under
different computational resources. However, the achieved final grasp is optimal
only in local sense.
The block diagram in the Figure 3.2 shows the data flow and the main elab-
24