Re: [Robotics Toolbox] Understanding of Gravload/RNE

[Peter Corke]

can you send me an m-file that builds your robot model.  I suspect the problem is more kinematic, that the robot is in fact not vertical.

peter

On 15/01/2013, at 9:51 PM, Nico Scheiffert wrote:

Dear all,

 

i currently doublecheck my own implementation of the RNE algorithm in Maxima (sym toolbox is not available) with the one in the toolbox for Matlab.

 

When I create a rotatable inverse pendulum with the toolbox, assign a mass of 2.8kg to each, first and second link (COG 0 -0.21 0 ; COG -0.2 0 0), and put the second link up, so that the bar is in equilibrium, I assume to get a gravload tau of [0 0], since there shouldn't be any moment.

 

In fact, I get a different tau: [1.0e-15*0 -0.3364]. OK, first value is, from my point of view, inaccuracy (should be 0), but the second value is far too much from zero.

Funny thing is, if you multiply the pi/2 joint position with e.g. 1.0002 , it reduces to 0.0017 which is much more towards zero. (Play a bit with the multiplication factor, you will see)

 

For me, it looks like a problem with handling PI? (using Win7 32bit, Toolbox rev. 8, Matlab 2010b)

Or do I have a bigger lack in understanding the dynamics?

 

 

alpha   A   theta   D  R/P
-1.570796 0.000000 0.000000 0.310000 R (std)
0.000000 0.400000 0.000000 0.000000 R (std)

 

 

--
You received this message because you are subscribed to the Google Groups "Robotics & Machine Vision Toolboxes" group.
To view this discussion on the web, visit https://groups.google.com/d/msg/robotics-tool-box/-/yLDcIfBL4hoJ.
To post to this group, send an email to robotics...@googlegroups.com.
To unsubscribe from this group, send email to robotics-tool-...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/robotics-tool-box?hl=en-GB.

[Nico Scheiffert]

Here we go

 

Thanks for your time

 

Nico

[Peter Corke]

This is code for a pretty ancient version of the toolbox, it wont run with the current version.  which version are you using?

To view this discussion on the web, visit https://groups.google.com/d/msg/robotics-tool-box/-/JFhvuW-C-bQJ.

To post to this group, send an email to robotics...@googlegroups.com.
To unsubscribe from this group, send email to robotics-tool-...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/robotics-tool-box?hl=en-GB.

<pendulum.m>

[Nico Scheiffert]

Robotics Toolbox for Matlab (release 8)

[Peter Corke]

Why not the current version?

To view this discussion on the web, visit https://groups.google.com/d/msg/robotics-tool-box/-/K9GMDUX28fgJ.

[Nico Scheiffert]

Nothing special, just used it some time ago on my private laptop and thought the basic algorithms wouldn't be touched. So I kept release 8.

 

I can update to release 9 and test it again.

[Peter Corke]

I suspect the algorithm hasn't changed much, just be easier for to test your script if it ran with current.

To view this discussion on the web, visit https://groups.google.com/d/msg/robotics-tool-box/-/aUUuqoynK1oJ.

[Nico Scheiffert]

No problem.

 

I will change the code to fit the new conventions and post it again.

[Nico Scheiffert]

New File, Release 9.7 - same result.

[Peter Corke]

Your robot is vertical for the case

>> test.gravload([0 -pi/2])

ans =

   1.0e-15 *

         0   -0.3364

and the result is effectively [0 0].  Did you perhaps misread this?  The 1e-15 is a common multiplier.

peter

To view this discussion on the web, visit https://groups.google.com/d/msg/robotics-tool-box/-/NS55UmJn63UJ.

[Nico Scheiffert]

Shame on me. Put the multiplicator only to the zero.

Sorry for bothering you.

 

Next time when using Matlab, I put an epsilon criteria for zero inside. ^^

 

Have a nice evening.

[Peter Corke]

I like a happy ending.

If you'd just cut and pasted your whole session into the first email it'd have been picked up quicker...

To view this discussion on the web, visit https://groups.google.com/d/msg/robotics-tool-box/-/YlQSlRSttiEJ.

[Nico Scheiffert]

I will do so if there is a next time (hope, it won't be).

Staring too long at the screen makes me sometimes blind for the obvious.

 

Currently creating geometrical and analytical Jacobian for a KUKA LBR. Try to combine a impedance control with a downhill-simplex self-learning algorithm for movement, stiffness and damping. Will see if there is any interesting output.

[Peter Corke]

Would you consider donating the Kuka model to RTB?  Essentially create a file mdl_kuka_lbr, similar style to all the other mdl_* files.

peter

To view this discussion on the web, visit https://groups.google.com/d/msg/robotics-tool-box/-/ab9FcrfkvKoJ.

[Nico Scheiffert]

When I'm finished, I will give you the complete piece of work (will take approx. 3 month, also have to check the NDA of Fraunhofer, that's why i build up everything by myself).

 

Currently, I estimated some dynamic parameters of the robot for a first rough model (cylindric volumes, COGs centred, mass aprox.).

According to your explanation in the old maple toolbox , I need to do create a basis parameter set and identify the parameter. Or, I can exploit the Fast Research Interface, since I can get an inertia matrix and all joint torques of every position from there. Or I use an adaptive control rule and fit the real model to my created rough model. We will see, which one is the easiest and fastest approach.

 

We'll be in touch.

[Nico Scheiffert]

Just as information:

 

Luckily, my CAS RNE Implementation produces the same results as your toolbox (compared with different configurations).

I like looking at a symbolic solution since it also helps understanding the inner behaviour much better.

[Peter Corke]

Good news, and thanks for the update.

If you have the Symbolic Toolbox for matlab you can use all the toolbox functions symbolically, generate inertia, gravity, corioilis terms etc.  See the FAQ.

peter

[Nico Scheiffert]

Not available. I have to do everything the "hard" way.

 

 

 

I already read some papers about base parameter and regressor building.

My idea would be moving the robot where all joints are stimulated, take the inertia matrix from the KUKA system in every trajectory point and do a multiple regression/Global Optimization. I'm not sure if this is the most elegant solution and it has the disadvantage not being used for other robots.

 

Do you have any suggestions regarding the parameter identification of the dynamics in general?

 

We could move to personal mail if you would like.

 

Best

Nico

[Peter Corke]

I'd love to have a Kuka model but I don't have the data to produce one.  Maybe try posting this in a new thread with a better heading.  Please consider donating it to RTB.

peter

On Tuesday, 26 February 2013 04:04:27 UTC+10, ravur...@gmail.com wrote:

can i get the any matlab mfile for kuka robot same as puma560 it will help me in doing my project

[Peter Corke]

quite a few: twolink, stanford, coil, ball, phantomx

On Thursday, February 28, 2013 4:15:59 AM UTC+10, ravur...@gmail.com wrote:

is there any mfile apart from pumarobot  which is done by u