Disturbance Force Application on Composite Bodies and Visualization in MuJoCo Simulations #2671

Kelly-234 · 2025-06-06T18:21:03Z

Kelly-234
Jun 6, 2025

Intro

Hi!

I am a student using MuJoCo to complete simulation tasks.

My setup

MuJoCo 3.2.6 (x64 Windows 11)

My question

I'm currently studying how my model behaves under external force disturbances and have created a simple model to familiarize myself with force application methods. I'd like to confirm several points:

When I apply a disturbance force to a body, is the force applied at the center of mass rather than the body's defined position (body pos)?

When bodies have parent-child relationships, is the external force applied at the parent body's center? Or is it applied to the composite center while maintaining fixed relative positions between bodies?

For example, in my model when I apply force to body0, I observe all three bodies moving together while maintaining fixed distances between them. I'm uncertain whether this occurs because:

The force is applied to the leftmost block specifically, or

The three blocks are treated as a single composite body with force applied at the center block

Additionally, how can I visualize disturbance forces similar to when dragging objects with the mouse in the simulation interface? I've attempted adding the following to my code but it doesn't seem to work.

        self.options = mujoco.MjvOption()
        mujoco.mjv_defaultOption(self.options)
        self.options.flags[mujoco.mjtVisFlag.mjVIS_CONTACTPOINT] = True
        self.options.flags[mujoco.mjtVisFlag.mjVIS_CONTACTFORCE] = True
        self.options.flags[mujoco.mjtVisFlag.mjVIS_TRANSPARENT] = True
        self.model.vis.scale.contactwidth = 0.1
        self.model.vis.scale.contactheight = 0.03
        self.model.vis.scale.forcewidth = 0.05
        self.model.vis.map.force = 0.3

Minimal model and/or code that explain my question

Model:

<mujoco>
    <option gravity="0 0 0"/>
    <worldbody>
        <light directional="true" diffuse=".8 .8 .8" specular=".2 .2 .2" dir="0 1 -1"/>
        <body name="cube0" pos="0 0 0">
            <joint type="slide" axis="1 0 0" damping="100"/>
            <joint type="slide" axis="0 1 0" damping="100"/>
            <joint type="slide" axis="0 0 1" damping="100"/>
            <geom name="cube" type="box" fromto="0 0 0 .1 0 0" size=".05" mass="1" rgba=".6 .8 .9 1"/>
            <body name="cube1" pos=".1 0 0">
                <joint type="slide" axis="1 0 0" damping="100"/>
                <joint type="slide" axis="0 1 0" damping="100"/>
                <joint type="slide" axis="0 0 1" damping="100"/>
                <geom name="cube1" type="box" fromto="0 0 0 .1 0 0" size=".05" mass="1" rgba=".3 .6 .8 1"/>
                <body name="cube2" pos=".1 0 0">
                    <joint type="slide" axis="1 0 0" damping="100"/>
                    <joint type="slide" axis="0 1 0" damping="100"/>
                    <joint type="slide" axis="0 0 1" damping="100"/>
                    <geom name="cube2" type="box" fromto="0 0 0  .1 0 0" size=".05" mass="1" rgba=".1 .3 .6 1"/>
                </body>
            </body>
        </body>
    </worldbody>
</mujoco>

Code:

import mujoco
import numpy as np
import mujoco.viewer
import matplotlib.pyplot as plt
from matplotlib.animation import FuncAnimation
import time

class CableDrivenArmController:
    def __init__(self, model_path):
        self.model = mujoco.MjModel.from_xml_path(model_path)
        self.data = mujoco.MjData(self.model)
        self.options = mujoco.MjvOption()
        mujoco.mjv_defaultOption(self.options)
        self.options.flags[mujoco.mjtVisFlag.mjVIS_CONTACTPOINT] = True
        self.options.flags[mujoco.mjtVisFlag.mjVIS_CONTACTFORCE] = True
        self.options.flags[mujoco.mjtVisFlag.mjVIS_TRANSPARENT] = True
        self.model.vis.scale.contactwidth = 0.1
        self.model.vis.scale.contactheight = 0.03
        self.model.vis.scale.forcewidth = 0.05
        self.model.vis.map.force = 0.3

    def apply_external_forces_to_cubes(self, force_magnitude, total_steps=1000):
        cube_names = ['cube2', 'cube1', 'cube0']
        cube_z_positions = {name: [] for name in cube_names}
        time_steps = []

        with mujoco.viewer.launch_passive(self.model, self.data) as viewer:
            for cube_name in cube_names:
                cube_id = self.model.body(cube_name).id
                step = 0
                while viewer.is_running() and step < total_steps:
                    self.data.xfrc_applied[cube_id] = [0, 0, force_magnitude, 0, 0, 0]
                    mujoco.mj_step(self.model, self.data)
                    viewer.sync()
                    time.sleep(0.001)
                    for cube_name in cube_names:
                        cube_z_positions[cube_name].append(self.data.body(cube_name).xpos[2])
                    time_steps.append(self.data.time)
                    step += 1
                self.data.xfrc_applied[cube_id] = [0, 0, 0, 0, 0, 0]

        # 绘图
        plt.figure(figsize=(10, 6))
        for cube_name in cube_names:
            z_positions = np.array(cube_z_positions[cube_name])
            plt.plot(time_steps, z_positions, label=f'Z Position of {cube_name}')

        plt.title('Z Position of Cubes under Z-axis Force')
        plt.xlabel('Time (s)')
        plt.ylabel('Z Position (m)')
        plt.grid(True)
        plt.legend()
        plt.tight_layout()
        plt.show()

if __name__ == "__main__":
    controller = CableDrivenArmController("e:\\asus\\Documents\\Study\\Seafile\\Python\\mujoco_model\\cable-driven-arm\\force_test.xml")
    controller.apply_external_forces_to_cubes(force_magnitude=5)

Confirmations

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

Answered by v-r-a

Jun 7, 2025

You are correct, we can visualise the perturbation forces. If you edit your code as follows, you can see the d.xfrc_applied forces:

        with mujoco.viewer.launch_passive(self.model, self.data) as viewer:
            with viewer.lock():
                viewer.opt.flags[mujoco.mjtVisFlag.mjVIS_TRANSPARENT] = True
                viewer.opt.flags[mujoco.mjtVisFlag.mjVIS_PERTFORCE] = True

I think this is not the way to go while using the passive viewer:

mujoco.mjv_defaultOption(self.options)
        self.options.flags[mujoco.mjtVisFlag.mjVIS_CONTACTPOINT] = True
        self.options.flags[mujoco.mjtVisFlag.mjVIS_CONTACTFORCE] = True
        self.options.flags[mujoco.mjtVisFlag.mjVIS_TR…

View full answer

v-r-a · 2025-06-06T20:03:07Z

v-r-a
Jun 6, 2025

Hi, I am a MuJoCo user. I will attempt to answer your queries!

As you have written correctly, the external force/torque will be applied only to/about the body's COM/inertia frame. You can check mj_applyFT for further flexibility.
If I understand correctly, you have three masses connected in series with three dampers. The first body is connected to the ground via three (sliding) dampers. You are getting the said behaviour due to action-reaction pairs of three dampers in series. Just like the net spring constant of three springs in series, each having stiffness k, would be k/3, the damping experienced by the last body is 100/3, hence freer movement. You can probably do a trial with a 1D system. You can also try the joint stiffness option.
Based on my understanding, you will have to draw the forces you wish to visualise. You can only visualise contact forces. (Also, parent-child body collisions are ignored, hence you won't see any contact forces unless the first and the third cubes collide.)

See #2467, i.e., the introductory tutorial for drawing objects in the scene.

Here is a function I use to draw arrows:

def draw_arrow(user_scene: mjc.viewer.Handle.user_scn,
                p1: np.ndarray, 
                dirn: np.ndarray, 
                linewidth: float = 0.05,
                colour: np.ndarray = np.array([1, 0, 0, 0.5])):
    """
    Draw an arrow in 3D space.
    """

    # We will add only one geom in this function    
    user_scene.ngeom += 1
    idx_geom = user_scene.ngeom - 1

    linesize = np.zeros(3)
    
    linesize[2] = np.linalg.norm(dirn)
    linesize[0] = linewidth
    linesize[1] = linewidth

    # Draw an arrow only if non-zero arrow length is specified
    if linesize[2] > 1e-6:
        myquat = np.array([1.,0.,0.,0.])
        mjc.mju_quatZ2Vec(myquat, dirn)
        Rmat = np.zeros(9)
        mjc.mju_quat2Mat(Rmat, myquat)

        mjc.mjv_initGeom(user_scene.geoms[idx_geom],
                        type=mjc.mjtGeom.mjGEOM_ARROW,
                        size=linesize,
                        pos = p1,
                        mat=Rmat,
                        rgba=colour)

You can use it in the simulation loop as follows:

...
# Draw arrow
viewer.sync()
viewer.user_scn.ngeom = 0 # to erase all the objects drawn, else they will pile up 
...

1 reply

Kelly-234 Jun 7, 2025
Author

Thank you for your helpful response!

Regarding the third point, while reviewing the official documentation at https://docs.mujoco.cn/en/stable/programming/visualization.html#perturbations, I noticed this passage:

"In addition to affecting the simulation, the perturbation object is recognized by the abstract visualizer and can be rendered. This is done by adding a visual string to denote the positional difference, and a rotating cube to denote the reference orientation of the selected body. The perturbation forces themselves can also be rendered when the corresponding visualization flag in mjvOption is enabled."

I'm uncertain whether this describes the same functionality you mentioned. If not, could you please advise on how to implement this force visualization?

Additionally, to confirm whether the external force I'm applying is continuous (as many tutorials emphasize clearing forces after application) or instantaneous, I conducted the following test:

Use data.xfrc_applied at every simulation step.

       with mujoco.viewer.launch_passive(self.model, self.data) as viewer:
            time.sleep(10)
            for cube_name in cube_names:
                cube_id = self.model.body(cube_name).id
                step = 0
                while viewer.is_running() and step < total_steps:
                    self.data.xfrc_applied[cube_id] = [0, 0, force_magnitude, 0, 0, 0]
                    mujoco.mj_step(self.model, self.data)
                    viewer.sync()
                    time.sleep(0.001)
                    for cube_name in cube_names:
                        cube_z_positions[cube_name].append(self.data.body(cube_name).xpos[2])
                    time_steps.append(self.data.time)
                    step += 1
                self.data.xfrc_applied[cube_id] = [0, 0, 0, 0, 0, 0]

2025-06-07.135837_20250607_18272769.mp4

Apply data.xfrc_applied only once when switching to the specified block.

        with mujoco.viewer.launch_passive(self.model, self.data) as viewer:
            time.sleep(10)
            for cube_name in cube_names:
                cube_id = self.model.body(cube_name).id
                self.data.xfrc_applied[cube_id] = [0, 0, force_magnitude, 0, 0, 0]
                step = 0
                while viewer.is_running() and step < total_steps:
                    mujoco.mj_step(self.model, self.data)
                    viewer.sync()
                    time.sleep(0.001)
                    for cube_name in cube_names:
                        cube_z_positions[cube_name].append(self.data.body(cube_name).xpos[2])
                    time_steps.append(self.data.time)
                    step += 1
                self.data.xfrc_applied[cube_id] = [0, 0, 0, 0, 0, 0]

2025-06-07.135931_20250607_18294575.mp4

Does this indicate that data.xfrc_applied[cube_id] = [0, 0, force_magnitude, 0, 0, 0] actually applies an instantaneous force? Or am I misunderstanding its usage?

v-r-a · 2025-06-07T21:39:12Z

v-r-a
Jun 7, 2025

You are correct, we can visualise the perturbation forces. If you edit your code as follows, you can see the d.xfrc_applied forces:

        with mujoco.viewer.launch_passive(self.model, self.data) as viewer:
            with viewer.lock():
                viewer.opt.flags[mujoco.mjtVisFlag.mjVIS_TRANSPARENT] = True
                viewer.opt.flags[mujoco.mjtVisFlag.mjVIS_PERTFORCE] = True

I think this is not the way to go while using the passive viewer:

mujoco.mjv_defaultOption(self.options)
        self.options.flags[mujoco.mjtVisFlag.mjVIS_CONTACTPOINT] = True
        self.options.flags[mujoco.mjtVisFlag.mjVIS_CONTACTFORCE] = True
        self.options.flags[mujoco.mjtVisFlag.mjVIS_TRANSPARENT] = True

I was speaking about drawing custom geometries. It is not really needed in this case. Pardon me for diverging from the point!

If you were to perform just the simulation without any visualisation, i.e., just mj_step() back to back, then q/d.xfrc_applied are not cleared by default. Last but one paragraph here has some comments on this.

But, if you were to use the passive viewer, then it gets cleared somewhere while the scene is updated. I am not able to find the exact line in the source code, but I can see that the perturbation forces are reset every time you do viewer.sync() or the corresponding update scene call in C. You can check the behaviour by tweaking your second code. I have added lines to print d.xfrc_applied

 with mujoco.viewer.launch_passive(self.model, self.data) as viewer:
            time.sleep(10)
            for cube_name in cube_names:
                cube_id = self.model.body(cube_name).id
                self.data.xfrc_applied[cube_id] = [0, 0, force_magnitude, 0, 0, 0]
                step = 0
                while viewer.is_running() and step < total_steps:
                    mujoco.mj_step(self.model, self.data)
                    print(self.data.xfrc_applied[cube_id])
                    mujoco.mj_step(self.model, self.data)
                    print(self.data.xfrc_applied[cube_id])

                    viewer.sync()
                    print("vis sync", step)
                    time.sleep(0.001)
                    for cube_name in cube_names:
                        cube_z_positions[cube_name].append(self.data.body(cube_name).xpos[2])
                    time_steps.append(self.data.time)
                    step += 1
                self.data.xfrc_applied[cube_id] = [0, 0, 0, 0, 0, 0]

You can see that d.xfrc_applied is cleared only after the first sync.

1 reply

Kelly-234 Jun 9, 2025
Author

I understand, the issue has been resolved. Thank you for your assistance.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Disturbance Force Application on Composite Bodies and Visualization in MuJoCo Simulations #2671

Uh oh!

{{title}}

Uh oh!

Replies: 2 comments 2 replies

Uh oh!

{{title}}

Uh oh!

Uh oh!

{{title}}

Uh oh!

Uh oh!

{{title}}

Uh oh!

Uh oh!

{{title}}

Uh oh!

Select a reply

Uh oh!

Disturbance Force Application on Composite Bodies and Visualization in MuJoCo Simulations #2671

Uh oh!

Kelly-234 Jun 6, 2025

Intro

My setup

My question

Minimal model and/or code that explain my question

Confirmations

Replies: 2 comments · 2 replies

Uh oh!

v-r-a Jun 6, 2025

Uh oh!

Kelly-234 Jun 7, 2025 Author

Uh oh!

v-r-a Jun 7, 2025

Uh oh!

Kelly-234 Jun 9, 2025 Author

Kelly-234
Jun 6, 2025

Replies: 2 comments 2 replies

v-r-a
Jun 6, 2025

Kelly-234 Jun 7, 2025
Author

v-r-a
Jun 7, 2025

Kelly-234 Jun 9, 2025
Author