Sliding box behavior #2633
Replies: 3 comments 14 replies
-
|
Hi! Can you please add an XML MRE that we can use as a reference for the discussion? Thanks |
Beta Was this translation helpful? Give feedback.
-
|
I did, but for some reason it doesn't show up. I can see it if I edit the post though. Let me investigate. Weird. |
Beta Was this translation helpful? Give feedback.
-
|
This is what I see. |
Beta Was this translation helpful? Give feedback.
-
|
I fixed it, take a look at the source. |
Beta Was this translation helpful? Give feedback.
-
|
You need to wrap XML with backticks, like code 🙂 |
Beta Was this translation helpful? Give feedback.
-
|
Thanks, noted for next time... |
Beta Was this translation helpful? Give feedback.
-
|
Alright the system does not let me upload XML files either but it lets me upload zipped XML files so here it is. |
Beta Was this translation helpful? Give feedback.
-
|
While this is being investigated, a couple recommendations for more stable sliding with this example I found by playing around with it:
|
Beta Was this translation helpful? Give feedback.
-
|
As it is often the case, reducing the time step leads to more stable sliding as well. More importantly, this simply doesn't look wrong. Why should we expect a box to slide smoothly on an inclined plane at arbitrarily large velocities? Have you ever seen this happen? Note that if I artificially decrease the timestep a lot (e.g. timestep="0.0002") then I do get a box sliding smoothly on an inclined plane at arbitrarily large velocities. Therefore, I think this is in fact the correct behavior (and indeed, what I see in other simulators). |
Beta Was this translation helpful? Give feedback.
-
|
Not having much luck with this forum. I'll need to redo my videos I guess. |
Beta Was this translation helpful? Give feedback.
-
|
It took me a while to generate videos that stay under the file size limit but here they are in case it helps. This is my basic "friction ramp" test, where the friction incrementally changes from 0.0 to 1.0 from one ramp to the next. First, here is the reference behavior I am looking for (in Jolt, but the behavior is the same in other simulators). FrictionRamp_Reference.mp4Second, here is the behavior in MuJoCo, described above. FrictionRamp_MuJoCo_Default.mp4Then, still MuJoCo, this is what I see if I change the solref parameters to (-100; -0.5). Not only friction seems to disappear, but the behavior after touching the ground is obviously not good. FrictionRamp_MuJoCo_Modified_Solref.mp4Finally, here is a setup where a heavy box with an initial linear velocity slides and collides against a light box, and then the opposite, a light box slides and collides against a heavy box. In green you have the behavior from Jolt. Then the behavior from PhysX. Then the unexpected behavior from MuJoCo. It is the first time I see jumping boxes in this test :) JumpingBoxes.mp4This is not urgent and it would be legitimate for this to be a side-effect of the contact model. I am however interested in suggestions how to improve this, if possible (by tweaking the settings, I cannot really reduce the timestep, as all simulators run with the same timestep in this framework). |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the follow up with videos, you are right, I now think my suggestion of lowering down the damping parameter in solref is probably not the right way to go. I noticed it needs high stiffness to prevent penetration which in turn makes the contacts sporadic and not a sliding friction behaviour like what we are looking for. |
Beta Was this translation helpful? Give feedback.
-
|
You do that by putting a in the model and then loading the keyframe at runtime. I tried that today and I am able to reproduce the behavior in the simulate app: <mujoco model="JumpingBox">
<statistic meansize="1"/>
<option timestep="0.016666"/>
<worldbody>
<light directional="true" pos="-50 -20 10" dir="5 2 -1" diffuse="1 1 1"/>
<geom type="plane" pos="9 0 5" size="30 10 .1" rgba="0 .8 0 1" friction="0.5"/>
<body name="sliding box" pos="30 0 6">
<freejoint/>
<geom type="box" size="1 1 1" friction="0.5"/>
</body>
</worldbody>
<keyframe>
<key name="initial_velocity" qvel="-20 0 0 0 0 0"/>
</keyframe>
</mujoco> |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Intro
Hello, I am interested in physics engines and starting to learn MuJoCo.
My setup
MuJoCo 3.2.2, C API, Windows
My question
I create a simple low-friction box on an inclined plane. The expected behavior would be that the box slides smoothly all the way to the bottom. The observed behavior is that the box starts jumping, no matter what options I try.
Best I could find so far was reducing the timestep, but even at 500 Hz the box ultimately starts jumping.
My questions:
(Note: the behavior is even more dramatic on a horizontal plane when the box is given an initial horizontal linear velocity, in which case you get a strong jump that does not look plausible at all. I could try that in code and I can post a video of it, but I do not know how to apply the initial velocity in the XML yet).
Not filled as a bug since maybe it's just me doing something wrong.
Thank you.
Minimal model and/or code that explain my question
Confirmations
Beta Was this translation helpful? Give feedback.
All reactions