[ros-users] gazebo: robot model dances

John Hsu johnhsu at willowgarage.com
Wed Sep 29 17:56:23 UTC 2010

Hi Ricarda,

This sounds like another modeling issue.  Please post your urdf model so we
can help you identify the potential problem.

While I don't have a good tool for debugging bad robot models, I'll try and
put together a list of things to check here.

Is your robot actuated?  If so, please check to see that the controllers are
outputting correct/reasonable commands (no near infinite accelerations as a
result).  This includes the safety controller if you are using any.

Meticulously double checking all masses and moi's.

Does reducing time step by factor of 100 help stabilizing things?  If so,
you should be able to identify the misbehaving joint/link by slowly
re-increasing the time step until you can see a single link/joint lead off
the "dance of instability".  Alternatively, break down your model into
pieces to identify the source of the problem.  The instability ("dance" or
"explosion") could be caused by a number of things, I'll list a few:

   - invalid mass, moment of inertias.
   - joint angle outside of joint limits (as in you previous model/email
   thread), incompatible joint limits.
   - unexpected collisions (possibly due to bad collision mesh).
   - time step size too large for your mass/moi ratios of connected links.
   If the physical system you are modeling is too stiff, then unfortunately you
   have to reduce time step size (you could potentially increase inner
   iteration counts or lower relaxation parameter, but that may not be
   sufficient sometimes).
   - too many redundant joints between the same bodies
   - incompatible constraints.
   - etc.

As you mentioned, changing mass values will affect the dynamics and
numerical properties of your physical system you're modeling.


On Wednesday, September 29, 2010, Ricarda Steffens <
steffens at informatik.uni-bonn.de> wrote:
> Hi everybody,
> after fixing problems such as "robot falling from sky", "robot explodes"
> and "robot explodes with help of a timed detonator" I now have the problem
> that my robot model "dances".
> The model spawns as it should be but then has a major tremor and falls on
> the floor still twitching.
> I multiplied the inertia matrix of the wheels by 100 which seems to solve
> the problem surprisingly. At least until I add the robot's arms.
> Then the arms and some other connected links, which were not twitching
> beforehand, have a small but noticable tremor. The tremor decreases but
> does not vanish when changing the mass value of the upper arm from 3.0 to
> 0.1.
> I did some trial-and-error stuff with the upper arm but could not find or
> fix the problem.
> So does anyone have an idea of how to solve this problem?
> Thanks,
> Ricarda
> _______________________________________________
> ros-users mailing list
> ros-users at code.ros.org
> https://code.ros.org/mailman/listinfo/ros-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ros.org/pipermail/ros-users/attachments/20100929/8b588e8c/attachment-0003.html>

More information about the ros-users mailing list