Steve Ims
PhD Student, Electrical Engineering
sdi@sun-valley.stanford.edu
Research
What's the significance of a flexible
manipulator?
Technically, all physical structures are flexible to some degree, but
we can make a (very good) engineering approximation that a manipulator
is rigid if its structural resonances are well above the bandwidth of
interest. For example, if a manipulator with a first structural
resonance at 400 Hz is being used for tasks that only require a
bandwidth of 10 Hz, then the flexibility can be neglected and the
manipulator modelled as a perfectly rigid structure. If we can
neglect the flexibility, then the position of the "working end" of the
robot can be inferred from simple geometry using the easy-to-measure
joint angles. In addition, if each joint is actuated then the
manipulator has the nice feature that all its degrees of freedom are
independently actuated. These two characteristics greatly simplify
the design of an automatic controller for the rigid manipulator.
However, if the rigid approximation is not suitable for a given
manipulator, then the flexibility must be addressed. As a result,
both modelling and design of an automatic controller become more
complicated.
What about the unknown payload?
As first shown in 1985 by Eric Schmitz, precise tip-position control
of a flexible manipulator can be achieved using the Linear-Quadratic
design methodology given an accurate system model. The dependence on
the system model is clear considering that the controller must apply
actuation at one end of the manipulator while trying to control
position of the other end, acting through the flexible manipulator.
When the manipulator picks up a payload, the dynamics of the new
system can be much different than those of the manipulator alone; if
the controller is not modified accordingly then poor performance or
instability can arise. In the case where the payload is unknown, the
affect on performance is of great concern. Further, additional modes
introduced by a dynamic payload can cause greater sensitivity to an
inaccurate model.
Hasn't this problem been looked at with adaptive
control?
Yes, many times.
Prof. Stephen Yurkovich at Ohio State University has led a
comprehensive research program in this area. In addition, much
work has been done within the ARL. Dan Rovner studied the adaptation
to an unknown tip mass by using deviations in the tip-position
trajectory from expected in order to determine the actual tip mass.
More recently, Larry
Alder developed an adaptive controller capable of responding to a
payload with one unknown dynamic mode by exploiting spectral
separation between the closed-loop modes and the payload mode. There
is also related work in the aerospace industry on wing-flutter
suppression.
Ok, so what's different here?
That's the trick. Tell you about it in a couple months...
Real Life
Stanford Aerospace Robotics Laboratory home page