Why min-time is necessary?

[snozawa]

@furushchev , @garaemon , @aginika
I have several questions for min-time in :angle-vector-duration:
https://github.com/jsk-ros-pkg/jsk_pr2eus/blob/master/pr2eus/robot-interface.l#L305

• Why min-time is necessary? In :angle-vector-duration, :max-joint-velocity is used. Isn't it enough?
• Why min-time is set to 1.0 by default? Sometimes old codes include :angle-vector smaller than 1.0 sec. Constant time 1.0 is too big for small joint difference.

[snozawa]

In addition, what is the best way to use :angle-vector less than 1.0[sec]?

[k-okada]

#121 (comment)
said

LGTM because

   - staro -> all the programs are old and it does not have scale option
   - hrp2 -> min-time=1.0 and scale=1 is enough

:anlge-vector is designed for beginner users to change robot posture and
for expert user who use motion planning or carefully prepared motion will
expect to use angle-vector-sequence.

◉ Kei Okada

On Sat, Dec 26, 2015 at 12:47 PM, Shunichi Nozawa notifications@github.com
wrote:

In addition, what is the best way to use :angle-vector less than 1.0[sec]?

—
Reply to this email directly or view it on GitHub
#203 (comment)
.

[snozawa]

:anlge-vector is designed for beginner users to change robot posture and for expert user who use motion planning or carefully prepared motion will expect to use angle-vector-sequence.

I see.
Default value for min-time is different between :angle-vector and :angle-vector-sequence,
but I understand the reason:
https://github.com/jsk-ros-pkg/jsk_pr2eus/blob/master/pr2eus/robot-interface.l#L379

expert user who use motion planning or carefully prepared motion will expect to use angle-vector-sequence.

In my case, I tested head look-at control by joystick controller, which worked on old OpenHRP2+rats system.
In the code, :angle-vector is send with time 200[msec] and with the :wait-interpolation-smooth.

[k-okada]

i see, are you using linear interpolator for this example? did it work
smoothly? what is the parameter for :wait-interpolation-smooth? what was
the minimum rate time for this example?

◉ Kei Okada

On Mon, Dec 28, 2015 at 1:54 PM, Shunichi Nozawa notifications@github.com
wrote:

:anlge-vector is designed for beginner users to change robot posture and
for expert user who use motion planning or carefully prepared motion will
expect to use angle-vector-sequence.

I see.
Default value for min-time is different between :angle-vector and
:angle-vector-sequence,
but I understand the reason:

https://github.com/jsk-ros-pkg/jsk_pr2eus/blob/master/pr2eus/robot-interface.l#L379

expert user who use motion planning or carefully prepared motion will
expect to use angle-vector-sequence.

In my case, I tested head look-at control by joystick controller, which
worked on old OpenHRP2+rats system.
In the code, :angle-vector is send with time 200[msec] and with the
:wait-interpolation-smooth.

—
Reply to this email directly or view it on GitHub
#203 (comment)
.

[snozawa]

Here is my code:
https://github.com/jsk-ros-pkg/jsk-ros-pkg-unreleased/blob/master/hrp2_controllers/hrp2_cmd_vel/src/hrp2-cmd-vel-common.l#L169-L186 (private repo)
Simplified code is like this:

(do-until-key
      (send *ri* :wait-interpolation-smooth 50 :head-controller)
      (send *ri* :angle-vector (send *robot* :angle-vector) 500 :head-controller 0 :min-time 0.0)
  )

I send :angle-vector 500[msec] and use :wait-interpolation-smooth with 50[msec].
It works on real HRP2 robot by setting min-time as 0 and I did not tune these time parameter so seriously.
I use hoff-arbib interpolation mode of SequencePlayer in hrpsys-base.

[k-okada]

what happens if you did not set :min-time 0 ?

◉ Kei Okada

On Mon, Dec 28, 2015 at 6:01 PM, Shunichi Nozawa notifications@github.com
wrote:

Here is my code:

https://github.com/jsk-ros-pkg/jsk-ros-pkg-unreleased/blob/master/hrp2_controllers/hrp2_cmd_vel/src/hrp2-cmd-vel-common.l#L169-L186
(private repo)
Simplified code is like this:

(do-until-key
(send ri :wait-interpolation-smooth 50 :head-controller)
(send ri :angle-vector (send robot :angle-vector) 500 :head-controller 0 :min-time 0.0)
)

I send :angle-vector 500[msec] and use :wait-interpolation-smooth with
50[msec].
It works on real HRP2 robot by setting min-time as 0 and I did not tune
these time parameter so seriously.
I use hoff-arbib interpolation mode of SequencePlayer in hrpsys-base.

—
Reply to this email directly or view it on GitHub
#203 (comment)
.

[snozawa]

Oh, it worked, but it is unexpected behaviour.

On real HRP2, I tried two commands:

(bench
(send *ri* :angle-vector (send *robot* :angle-vector) 1000 :head-controller)
(send *ri* :wait-interpolation :head-controller))
;; time -> 1.46333[s]

This output is expected.
1.46[s] is more than 1000[msec], so :wait-interpolation successfully wait for finishing :head-controller interpolation.

However,

(bench
(send *ri* :angle-vector (send *robot* :angle-vector) 1000 :head-controller)                                      
(send *ri* :wait-interpolation-smooth 10 :head-controller))
;; time -> 0.015762[s]

In this case, expected output is like ;; time -> 0.990[s] or something.
Returning of :wait-interpolation-smooth seem to be too early.

The euslisp test code (https://github.com/jsk-ros-pkg/jsk_pr2eus/blob/master/pr2eus/test/default-ri-test.l#L35) tests :wait-interpolation-smooth, but it is simulation mode.
Which part checks the behaviours of :wait-interpolation and :wait-interpolation-smooth of non-simulation mode?