Modeling badminton racket as an elastic rod using cable #1386

alex1yaremchuk · 2024-02-01T09:24:42Z

alex1yaremchuk
Feb 1, 2024

Hi,

I'm an amateur researcher and I'm trying to use MuJoCo for modeling a badminton racket (later with an arm).
I'm not a physicist but I'm a programmer.

I'm aiming at modeling a racket rotating around a wrist for the beginning.
I want to compare speeds of a rigid stick and flexible rod ends and therefore measure the whipping effect.
Later I wanted to model shoulder-elbow-wrist system and find optimal swing using reinforcement learning.

I'm looking for some help with understanding how to simulate an elastic rod of approximate badminton racket characteristics.
I was successfull at rigging up a flexible stick on a hinge which can be rotated using an actuator.

But when I try to play with it, f.i. make it thinner it becomes unstable.
I have to use quite a small timestep to make it work.

Questions:

Can I have a general advice - am I doing right or my simulation can be made more robust?
I need it to be quite effective as I suppose to use it for RL later - so it should calculate quickly.
Can I have an advice on bringing my rod closer to real badminton racket on physical characteristics?

Here is a notebook I'm experimenting with (section Flexible rod based on cable) -
https://github.com/google-deepmind/mujoco#asking-questions

Here is a model from this notebook:

minimal XML

<mujoco model="smash">
  <option integrator="RK4" timestep="0.00002"/>
  <extension>
    <plugin plugin="mujoco.elasticity.cable"/>
  </extension>
  <asset>
    <texture name="grid" type="2d" builtin="checker"
      width="300" height="400" />
    <material name="grid" texture="grid" texrepeat="8 8"
      reflectance=".2" />
  </asset>

  <compiler autolimits="true"/>

 <size memory="200M" />


  <worldbody>
    <light diffuse="10.5 10.5 10.5" name="top" pos="0 0 6" dir="0 0 -1"/>
<!--    <camera name="cam1" pos="0 -7 4" euler="80 0 0" /> -->
<!-- doesn't work   <camera name="cam1" pos="0 -7 4" axisangle="1 1 1 0 " /> -->
    <camera name="cam1" pos="0 -7 4" zaxis="0 -1 0.2"  />
    <!-- <geom size="10 10 .1" type="plane" material="grid" /> -->
    <geom size="10 10 .1" type="plane" rgba="0 .9 0 1" />

    <body name="shoulder_body" >
      <joint name="swing" type="hinge" axis="0 1 0" pos="0 0 3"/>
      <composite type="cable" curve="s" 
        count="20 1 1" size="2" offset="0 0 3" initial="none">
        <plugin plugin="mujoco.elasticity.cable">
          <!--Units are in Pa (SI)
          <config key="twist" value="1e7"/>
          <config key="bend" value="6e6"/>
          <config key="vmax" value="0.05"/> -->

          <config key="twist" value="1e9"/>
          <config key="bend" value="1e8"/>
          <config key="vmax" value="0.05"/>

        </plugin>
        <joint kind="main" damping=".015"/> 
        <geom type="capsule" size=".05 .5" rgba="1 0 0 1" 
          condim="1"/>
      </composite>
    </body>
  </worldbody>
<!--  <keyframe>
    <key name="swinging" qvel="20"/>
    <key name="steady" qvel="0"/>
  </keyframe> -->
  <actuator>
    <motor name="swing_motor" joint="swing" ctrlrange="-1000 1000"/>
  </actuator>
</mujoco>

Here is a video of simple simulation of applying force for a second and opposite force later.

flexible.rod.on.a.hinge.mp4

All feedback is appreciated!

Balint-H · 2024-02-01T16:59:30Z

Balint-H
Feb 1, 2024

I was able to get the rod to behave in a similar way at 100 times larger timestep with the following attributes inside the joint definition of the composite body:

<joint kind="main" damping="20" stiffness="2000" armature="0.01"/>

If it's too wobbly, bump up the stiffness to 10000<.

I'm not sure how specifying stiffness there interacts with the twist and bend parameters of the cable though. I would have though the stiffness would be determined by them, but specifying both cable properties and stiffness seems to be necessary for a stable simulation. Armature in general helps stabilize exploding joints.

Also, the implicit integrator seems to be more stable than RK4 here.

7 replies

Balint-H Feb 4, 2024

Heya when you say absolutely rigid do you mean no bend at all? Because that is of course very easy.

Decreasing the number of segments will also likely decrease the wobbliness and make it more stable.

Also play a bit with the armature arguement, I added that to prevent exploding, but too much can add unneeded inertia to the rod, also making it wobbly.

alex1yaremchuk Feb 4, 2024
Author

Hey! thank you for your prompt reply, I'll give it a try.

Yeap, I thought of having a scale from absolutely rigid one to a flexible stick.
I had in mind showing

impact of whipping effect on the resulting speed (I'd like to base whipping effect on absolutely rigid one)
dependency on the motor power (player strength) and the optimal flexibility of the racket corresponding to it.

alex1yaremchuk Feb 4, 2024
Author

Yeah, it becomes unstable with lower armature..

alex1yaremchuk Feb 4, 2024
Author

As for number of segments - probably I don't need 20.

But I'm not clear about that - initially I thought that the more parts - the closer to real rod it will be.
(I don't care about smooth rendering but only about adequate modelling)

What I'm expecting from the flexibility model is that with increasing flexibility the rod is going to be bend more in a bow as a whole.
but what I see now - it behaves like a rope or a sausage.

I have some doubts - did I correctly choose cable, maybe I need a rope for such a lengthty object?

Balint-H Feb 7, 2024

@quagla can you advise how the elasticity plugin could be used to achieve this kind of high stiffness bending?

alex1yaremchuk · 2024-02-07T09:18:55Z

alex1yaremchuk
Feb 7, 2024
Author

Another idea - may be Mujoco is not the right emulation framework for my needs as I don't need any contacts or some other complicated niceties.
I just need a rod simulation (and later several joints arm simulation)

0 replies

quagla · 2024-03-06T16:58:39Z

quagla
Mar 6, 2024
Collaborator

Discussed and solved in #1399.

0 replies

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Modeling badminton racket as an elastic rod using cable #1386

{{title}}

Replies: 3 comments 7 replies

{{title}}

{{editor}}'s edit

{{editor}}'s edit

{{title}}

{{editor}}'s edit

{{editor}}'s edit

{{title}}

{{title}}

{{title}}

{{title}}

{{title}}

{{title}}

Select a reply

Modeling badminton racket as an elastic rod using cable #1386

alex1yaremchuk Feb 1, 2024

Replies: 3 comments · 7 replies

Balint-H Feb 1, 2024

Balint-H Feb 4, 2024

alex1yaremchuk Feb 4, 2024 Author

alex1yaremchuk Feb 4, 2024 Author

alex1yaremchuk Feb 4, 2024 Author

Balint-H Feb 7, 2024

alex1yaremchuk Feb 7, 2024 Author

quagla Mar 6, 2024 Collaborator

alex1yaremchuk
Feb 1, 2024

Replies: 3 comments 7 replies

Balint-H
Feb 1, 2024

alex1yaremchuk Feb 4, 2024
Author

alex1yaremchuk Feb 4, 2024
Author

alex1yaremchuk Feb 4, 2024
Author

alex1yaremchuk
Feb 7, 2024
Author

quagla
Mar 6, 2024
Collaborator