Two geometries have penetrated, but constraint force is 0 #2694

ppap36 · 2025-06-19T14:10:47Z

ppap36
Jun 19, 2025

Intro

Hi!

I am a student at BIT, I use MuJoCo for my research on compliant capture control.

My setup

Mujoco 3.3.1

My question

I am trying to validate the constraint acceleration formula.

I am trying to upload two boxes, and let one box drop on another box by gravity. The falling box first penetrates the stationary box, then bounces upward, continues to fall under gravity, penetrates again, and finally stabilizes. I use geomDistance to obtain the minimum distance dist between the two objects. I thought that as long as dist < 0 (i.e., penetration occurs), there must be a constraint-based contact force between them. However, in the code I provided, when I output the distance and contact force from timestep 1088 to 1098, although dist < 0, the contact force is 0. I don’t understand why this happens.

Minimal model and/or code that explain my question

If you encountered the issue in a complex model, please simplify it as much as possible (while still reproducing the issue).

import mujoco
import numpy as np

mjcf = """
<mujoco>

    <option gravity = "0 0 -4"/>
    <worldbody>

    <geom name="floor" type="plane" pos="0 0 0" size="10 10 0.05"/>
    <body name="cube1" pos="0 0 0.5">
        <geom name="box1" type="box" mass="10" size="0.5 0.5 0.5" rgba="1 0 0 0.5"
              solref="0.02 1" solimp="0.99 0.99 0.001 0.5 1" contype="1" conaffinity="1"/>
        
    </body>

    <body name="cube2" pos="0 0 10.25">
    <freejoint/>
        <geom name="box2" type="box" mass="1" size="0.25 0.25 0.25" rgba="0 0 1 0.5"
              solref="0.02 1" solimp="0.99 0.99 0.001 0.5 1" contype="1" conaffinity="1"/>
        
    </body>    

    </worldbody>
</mujoco>
"""

model = mujoco.MjModel.from_xml_string(mjcf)
data = mujoco.MjData(model)
n = 0

mujoco.mj_resetData(model, data)

dist = []
qvel = []
qacc = []
force = []
qfrc_constraint = []
while n<10000:
    mujoco.mj_step(model, data)


    qvel.append(data.qvel[2])
    qacc.append(data.qacc[2])
    qfrc_constraint.append(data.qfrc_constraint[2])

    geom1_id = 1
    geom2_id = 2


    fromto = np.zeros(6, dtype=np.float64) 


    distmax = 10.0  
    dist0 = mujoco.mj_geomDistance(model, data, geom1_id, geom2_id, distmax,fromto)
    dist.append(dist0)
    if data.ncon > 0:
        force0 = np.zeros(6)
        mujoco.mj_contactForce(model, data, 0, force0)
        force.append(force0[0])
    else:
        force.append(0)

    n += 1


force = np.array(force)
qacc = np.array(qacc)
qvel = np.array(qvel)
dist = np.array(dist)

start = 1088
end = start + 10
for i in range(10):
    print(dist[start:end][i],qvel[start:end][i],qacc[start:end][i],force[start:end][i],qfrc_constraint[start:end][i])

Confirmations

I searched the latest documentation thoroughly before posting.
I searched previous Issues and Discussions, I am certain this has not been raised before.

v-r-a · 2025-07-05T16:36:41Z

v-r-a
Jul 5, 2025

Hi,

I thought that as long as dist < 0 (i.e., penetration occurs), there must be a constraint-based contact force between them.
This would be true only in a static case.

I tweaked your model so that the boxes start off as penetrated.
Then applied an initial upward force.
You can see that with a large enough initial upward force, there is no contact force even with large penetration. While I have little understanding of the contact model, I can say that intuitively this is the case where you press a material and pull back your finger faster than the material can recover.

You can increase the damping in solref to more than 1 and see this bounce effect fade away; increasing the damping won't let the box gain enough upward acceleration. You also seem to have increased the solimp[0] to a higher value that the default one, this leads to higher contact forces at the start of collision.

mjcf = """
<mujoco>

    <option gravity = "0 0 -4"/>
    <worldbody>
    <geom name="floor" type="plane" pos="0 0 0" size="10 10 0.05"/>
    <body name="cube1" pos="0 0 0.5">
        <geom name="box1" type="box" mass="10" size="0.5 0.5 0.5" rgba="1 0 0 0.5"
              solref="0.02 1" solimp="0.99 0.99 0.001 0.5 1" contype="1" conaffinity="1"/>        
    </body>

    <body name="cube2" pos="0 0 1.1">
    <freejoint/>
        <geom name="box2" type="box" mass="1" size="0.25 0.25 0.25" rgba="0 0 1 0.5"
              solref="0.02 1" solimp="0.95 0.99 0.001 0.5 1" contype="1" conaffinity="1"/>        
    </body>    

    </worldbody>
</mujoco>
"""

model = mujoco.MjModel.from_xml_string(mjcf)
data = mujoco.MjData(model)
n = 0

mujoco.mj_resetData(model, data)
data.xfrc_applied[2][2] = 1000.
mujoco.mj_forward(model, data)

# same as your code
# print initial 50 time steps or so

0 replies

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Two geometries have penetrated, but constraint force is 0 #2694

Uh oh!

{{title}}

Uh oh!

Replies: 1 comment

Uh oh!

{{title}}

Uh oh!

Select a reply

Uh oh!

Two geometries have penetrated, but constraint force is 0 #2694

Uh oh!

ppap36 Jun 19, 2025

Intro

My setup

My question

Minimal model and/or code that explain my question

Confirmations

Replies: 1 comment

Uh oh!

v-r-a Jul 5, 2025

ppap36
Jun 19, 2025

v-r-a
Jul 5, 2025