Takahashi et al:


The authors present a 3 mode system to control a hand for
grasping (position, force-constrained position, and force) which operate
respectively in regions 
far from contact, near contact and after grasp. They also describe a 
slip sensor which allows them to determine required grasp
forces at run time. 

The paper spends too much time on topics which do not appear to be new
contributions (e.g. disturbance observer, coordinate transforms, etc.).
Instead, the paper would be more interesting, and more clear, if more
space were devoted to explaining the slip estimator.

The authors first describe the low level of the control. Despite the
length of explanation, this reviewer was only able to understand it
fully after consulting some of the references listed by the authors. In
retrospect, it appears that the only novel contribution is the addition
of a pseudo-differentiator block and adjusting the cutoff frequency for
the disturbance rejection itself. The authors also have a confusing
mixed notation: F_hat_dis in Figure 2 versus tau_hat_dis in eqn 2 (Are
these not the same thing?)

The control mode switching idea seems good although not entirely novel
(others have used various mode switching algorithms, although typically
involving 2 modes rather than 3 (e.g. Phase/Event framework for
dexterous manipulation). It would be useful to know more about what the
criteria are for switching, and how robust the indicator functions are.
Also, how are transient errors treated at the startup of new control
phase? The authors claim that their switching mode control allows for
"smooth and quick" transitions but do not provide evidence for
understanding why this must be so.

The graphs in fig 9 and 10 about the performance of this controller are
hard to read.

The tactile sensor and slip detection section is the hardest to follow.
The authors present several detailed equations that are meant to compute
a value for "suppression of the desired grasp force" with little to no
explanation beyond "this value is a constant, this value is the force
from a sensor...etc". Several items are not fully explained. Also, there
is some inconsistency in the notation. For example, in Section IV.A the
authors say that "i" is the time step; later it is "k". It appears they
base the slip calculation on the center of pressure using an array of
capacitive sensors at the fingertip. Presumably they are using the joint
kinematics to distinguish between rolling and sliding motion? Their
final results from this section test their grasp force calculation based
on slip extractors against a constant force grasp. But this seems to be
a rather special case because they picked a slippery cube and give their
constant grasp force an unrealistically low value.

There is considerable time spent explaining the three modes that are being switched between, but there is no in depth discussion about the criteria for switching. Figures 3,4, and 5 present semi-redundant information concerning the control modes and the need of switching, especially Figures 3 and 4.

The controller gains notation used in the eqns 10 and 12 does not match the notation used in the block diagram in Figure 5, K_d = K_v?**

In section IV part B and C, there are many constants defined with a value without any explanation of how the value was determined. In figure 12, it is difficult to distinguish between the different values being plotted in the force versus time plot. Figure 13 has a caption that reads "Experimental result 10-300(g)", but the plot shows values 0g, 100g, 200g and 300g. So should the caption read "Experimental result 0-300(g)"? During the discussion of the results shown in Figure 13, the authors mention the values reach a steady state, but to the reader without any definitions of "steady state" it is not straight-forward to distinguish the steady state value. In Figure 14, it should be more clearly labeled that the mass includes both the object's original mass and the additional mass, because previous results are discussed with respect to only the additional mass. To be more consistent, the authors should describe the mass of the experiment the same throughout all figures.

In section V. Conclusion, there is a spelling error in the third sentence of the first paragraph, "resutls" -> "results" and in the fourth sentence, "respons" -> "response".