Human-machine Cooperative Manipulation with Vision-based Motion Constraints - Visual Servoing via Advanced Numerical Methods - page 52

of medical interventions. Our goal is to create mechanisms that are neither au-
tonomous, nor purely passive. Rather, our intent is to create mechanisms that se-
lectively provide cooperative assistance to a surgeon, while allowing the surgeon to
retain ultimate control of the procedure.
In our recent work, we have focused on developing assistance methods for mi-
crosurgery. Here, the extreme challenges of physical scale accentuate the need for
dexterity enhancement, but the unstructured nature of the tasks dictates that the hu-
man be directly “in the loop”. For example, retinal vein cannulation [31] involves
the insertion of a needle of approx. 20-50 microns in diameter into the lumen of
a retinal vein (typically 100 microns in diameter or less) 1 . At these scales, tactile
feedback is practically non-existent, and depth perception is limited to what can be
seen through a stereo surgical microscope. In short, such a procedure is at the limit
of what is humanly possible in conventional surgical practice.
Given the scale of operation, the most obvious need is to increase the precision of
human motion, ideally without slowing or limiting the surgeon. In recent work [19,
20, 18], we have begun to develop assistant methods that are based on manipulating
the apparent compliance of tools simultaneously held by both the surgeon and a
robot. Intuitively, if a tool is extremely stiff, then it is easier to achieve high precision
of motion, and to remove tremor. Conversely, low stiffness makes it possible to
perform large-scale “transport” motions.
Although they increase absolute precision, purely isotropic compliances cannot
take advantage of natural task constraints to provide structured assistance. For ex-
ample, when placing a needle into the lumen of a blood vessel, the natural mode
of assistance would be to stabilize the needle in the lateral directions, but permit
relatively free, quasi-static positioning along the needle axis.
In this chapter, we specifically focus on the use of anisotropic compliances as
a means of assistance. In previous work we have related these anisotropic compli-
ances to the notion of virtual fixtures [21, 28]. Virtual fixtures, like the real thing,
provide a surface that confines and/or guides motion. We initially describe how vir-
tual fixtures can be produced as a generalization of previous work in [1, 2, 24], and
then turn to the problem of deriving vision-based virtual fixtures. Through example,
we show how visual servoing algorithms for one camera [3] and two camera [8, 10]
systems can be translated into virtual fixtures. As a result, much of the previous
literature on visual servoing can be applied to the problem of human-machine co-
operative manipulation. Finally, we describe several applications, both surgical and
nonsurgical, that we have developed around virtual fixtures of the type introduced
in this chapter.

4.2 Virtual Fixtures

Our work has been motivated by the Johns Hopkins University (JHU) steady hand
robot (SHR), and, in particular, the assistance paradigm of direct manipulation it was
designed for [18, 19, 30]. Briefly, the JHU SHR is a 7 degrees of freedom (DOF)

1

As a point of reference, a human hair is typically on the order of 80 microns in diameter.