Information Technology Reference
In-Depth Information
local minima [49]. An alternative framework to deal with unmodelled uncertainties
is to retreat the robot/camera and/or re-plan quickly when encountering violation in
a constraint. A variable zooming technique was suggested by [38] to bring the tar-
get back within the visibility range if occluded by an obstacle. This zooming effect
can also drastically improve the performance of the underlying image-based visual
servoing technique by reducing the measurement noise in fixed-size objects viewed
by a camera from distance.
Comport et al. [15] proposed an augmented reality approach for visual servoing.
Although their approach mainly focuses on camera pose estimation by means of a
virtual visual servoing method, but this can be extended to scenarios in which some
feature points on the target may go out of sight temporarily,
e.g.
due to unmod-
elled uncertainties. An augmented reality approach can then be utilized to virtually
position the missing feature points in the image based on rudimentary information
obtained from other objects in a scene cluttered with known features,
i.e.
straight
edges,
etc
. In this case, the target is used as the primary object for visual servoing
while other image features can contribute to the pose estimation, and eventually to
the servoing task, when a finite number of feature points fall off the cameras field
of view.
Nonlinear model predictive control strategies have been proposed to account for
uncertainties in planned trajectories in visual servo control loop as well,
e.g.
[57].
Systems' parameters would be corrected beyond a temporal receding horizon (
i.e.
,
the time span during which the optimal control action is computed and executed)
after each iteration. The discrepancy between the predicted system's behavior based
on the computed control action and that in real implementation is then used to fur-
ther correct the estimates of the system's parameters. The time required to estimate
these parameters via a nonlinear optimization technique must be way shorter than
the receding horizon in which this optimization is carried out. Otherwise, the appli-
cability of this technique for real-time scenarios would be questionable. Developing
a guideline for selecting the optimal size for the receding horizon for robust visual
servoing in real time remains an open research area.
Robustness with respect to calibration errors in terms of the tracking error bound-
edness along the planned trajectories has been considered in [53]. Given a user
defined bound on the tracking error, they propose a control strategy to modulate
control gains and/or the desired tracking velocity to guarantee error boundedness.
Through the proposed velocity modulation technique, one could use low control
gains while keeping the tracking error bounded. While this technique and those
mentioned above, to some extent, are expected to account for the deviations from
the planned trajectories in the image space, the deviations from the physical space
trajectories can cause robot/physical constraints violations. The above mentioned
local strategies for accounting deviations from planned path are either prone to lo-
cal minima or not general enough to account for all types of constraints (and the
related uncertainties), in particular robot/physical constraints. Hence, there is need
for taking the uncertainties into account in a global as well as general manner at the
planning stage.
Search WWH ::
Custom Search