Equality constraint maxing out

[M8e0]

Intro

Hi!

I am an undergrad student , I use MuJoCo to simulate pick and place movements using delta robots.

My setup

Mujoco 3.3.3 using python API on windows 11

My question

I am facing a problem with the soft equality constraints.

No matter the parameters I put in the model, the sites under the equality constraints are torn away when the robot is moving, thus making pick and place movements impossible because of the lack of precision.

Is there a way to make the equality constraint behave as a hard constraint ?

The workaround I found is to attach a mocap cube to the end-effector of the robot, and move it. sadly this approach shuts the door on using inverse kinematics to control the robot.

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).

parameters of the equality constraints :

<connect site1="mountingplate" site2="cube_origin_site" solref="0.0004 0.4" solimp="1 1 0.1 0.1 5"/>
    <connect site1="closing_rotule2_1" site2="closing_rotule2_2" solref="0.002 1" solimp="1 1 0.0001 0.001 30"/>
    <connect site1="closing_rotule3_1" site2="closing_rotule3_2" solref="0.002 1" solimp="1 1 0.0001 0.001 30"/>
    <connect site1="closing_rotule4_1" site2="closing_rotule4_2" solref="0.002 1" solimp="1 1 0.0001 0.001 30"/>
    <connect site1="closing_rotule5_1" site2="closing_rotule5_2" solref="0.002 1" solimp="1 1 0.0001 0.001 30"/>
    <connect site1="closing_rotule6_1" site2="closing_rotule6_2" solref="0.002 1" solimp="1 1 0.0001 0.001 30"/>

Code:

import mujoco
import mujoco.viewer
import numpy as np
from pathlib import Path
from position_control import EffectorMotionController

# Get correct directory 
robotfile = Path(__file__).parent.parent / "assets/robotLib/DeltaX3/robot.xml"

# Load model and data
model = mujoco.MjModel.from_xml_path(str(robotfile))
data = mujoco.MjData(model)


body_id = mujoco.mj_name2id(model, mujoco.mjtObj.mjOBJ_BODY, "mocap_cube")
mocap_id = model.body_mocapid[body_id]

controller = EffectorMotionController()
controller.generate_movement()

# Run simulation
with mujoco.viewer.launch_passive(model, data) as viewer:
    while viewer.is_running(): 
               
        pos_and_orientation = controller.get_next_position_and_orientation()
        
        data.mocap_pos[mocap_id] = pos_and_orientation[:3]
        data.mocap_quat[mocap_id] = pos_and_orientation[3:]
        
        mujoco.mj_step(model, data)
        viewer.sync()

Confirmations

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