Reviewer 1 of ICRA 2020 submission 720

Comments to the author
======================

The paper proposed a teleoperation system that allows for
complete control over the high degree-of-actuation robotic
system by only observing the bare human hand.  There are no
manipulators or gloves in the master system, but only
vision sensors capture the motion of the human hand.  Based
on the image processing, target joint angles are generated
in real-time.  The paper shows the proposed system is able
to realize highly complicated tasks.

How did the authors determine the value of the threshold
epsilon?

The authors should discuss the results of the tasks in more
detail.  For example, why the success rate of Brick Gaiting
task was so low?  What can be the hidden problem here?

Although the success rate of the Container task is 100%,
the time it took to complete the trial is very long.

Why the pilots gave up the Brick Gaiting task so early? 
They could have taken more time to get a better success
rate.  It seems this is the only task that explicitly
requires to make rotation motion.  Does the result have
something to do with this movement?

The mean completion time should be divided by the
completion time when we normally do the task for
normalization.	By evaluating the time in this ratio, we
can compare the tasks to see which task was easy or not. 

When considering to lower the cost further, we can also use
a three-fingered robot hand.  Does the difference in the
number of the finger dramatically affect the success rate? 

I want to know whether detecting the shape of the object
gives a positive effect to accomplishing the tasks.


Comments on the Video Attachment
================================

The video properly shows the proposed method can realize
complicated tasks.


Reviewer 2 of ICRA 2020 submission 720
  
Comments to the author
======================

 In this research, a low-cost, depth-based teleoperation
system, DexPilot, is developed that allows for complete
control over the full 23 DoA robotic system by observing
the bare human hand. DexPilot enables operators to solve
manipulation tasks that go beyond simple pick-and-place
operations. The DexPilot system is experimentally evaluated
across physical tasks that test precision and power grasps,
prehensile and non-prehensile manipulation, and finger
gaiting. The performance is measured through speed and
reliability metrics.
 This paper is well organized and is well motivated. The
reviewer recommends accepting for publication in ICRA2002.
Comments to the authors are described below.

* The authors need to mention or give a quantitative
consideration for the operability of DexPilot. In the upper
right of Fig. 1, the robot hand holds the cup even though
the operator's finger is completely closed. Therefore, it
is necessary to consider the effect of this difference on
the operator's feeling of operation. Also, it is required
to mention whether this situation is appropriate for a
teleoperation system.
* The figure on the first page should be placed at the
bottom of the page, not the top.
* If there are multiple reference numbers, they should be
combined using a hyphen (e.g., [8-11] or [8]-[11]). Please
check the manuscript format for the presentation in
ICRA2020.
* Section for the conclusion of this manuscript should be
given for the readability.
*Numbers for equations on page 4 should be given.


Comments on the Video Attachment
================================

This video well describes the proposed method.