Use the generated model in Gazebo

[diegoferigo]

In this issue I want to track and recap the efforts done (mainly by @nunoguedelha) and what's missing in the generation chain to let these model work in Gazebo.

• Creating a first version that simulates (URDF model generation with all the distributed inertial sensors #49). These changes for now have been applied manually to the icub-models repository (Updated models after icub-model-generator PR 49 icub-models#2)
• Install the model.config for all the robots (Install all the generated models  icub-models#3)
• Enable all the inertial sensors (wrapped in <sensor> tag). Right now gazebo crashes without giving any error (Gazebo crashes without throwing any error if we add the accelerometers to the URDF #54)
• Handle the name of the robot (Allow changing the robot name simmechanics-to-urdf#24)
• Include the snippet at the bottom (Add feet collision (bounce) and friction parameters. #55, Gazebo crashes without throwing any error if we add the accelerometers to the URDF #54, Reorder URDF file elements for avoiding sdformat issue 164 simmechanics-to-urdf#25)
• Handling the small inertias. Two ways for this: increasing the numbers that are extremely low, ideally in an automatic way, or exploiting the implicit damping like reported in Generate gazebo models with implicit damping enabled #24. Solved by Add iCubGazeboV2_5 model  #71.
• Integrate FT sensors plugins for Gazebo (Add parsing of xml blobs in FT sensor gazebo elements simmechanics-to-urdf#32, Add the missing FT sensors plugin info in the yaml config files #72)
• Allow using generated models within the composed models in icub-gazebo-wholebody repo (initial PR Allow using the configurations shipped in this project specifying the robot model name robotology-legacy/icub-gazebo-wholebody#12 and fix Fix/icub model config and fixation to world robotology-legacy/icub-gazebo-wholebody#15)
• Spawn the auto-generated iCub raw models in the composite Gazebo models (iCub fixed, iCub on seesaw, ...) (Spawn the auto-generated iCub raw models in the composite Gazebo models (iCub fixed, iCub on seesaw, ...) robotology-legacy/icub-gazebo-wholebody#16)
• Fix the model generation chain in order to be able to insert directly in Gazebo the URDF models from the repo icub-models (such models have a default pose and joint configuration and appear hovering 1mm above the ground) => Raw iCub model from icub-models repo cannot be used directly #79
• install configuration files only for iCubGazeboV2_5, as all other models are expected not to work (as per @traversaro recommendation) => install configuration files only for the supported models (iCubGazeboV2_5) icub-models#11

Other pending issues imported from robotology/icub-models#7:

• Both the wrist yaw joints go in hardware fault. Apparently the position control is unstable and it brings the joints toward the limits => In Gazebo both the wrist yaw joints go in hardware fault #74.
• The meshes in the foot should be substituted with a simpler primitive shapes (box or four spheres) => Change feet collision meshes to a simpler primitive shape #77

cc @nunoguedelha @traversaro @DanielePucci

---

<gazebo reference="l_foot">
    <collision>
        <surface>
            <bounce>
                <restitution_coefficient>0</restitution_coefficient>
                <threshold>100000</threshold>
            </bounce>
        </surface>
    </collision>
    <maxContacts>4</maxContacts>
    <kp>18000000</kp>
    <kd>100</kd>
    <maxVel>100</maxVel>
    <minDepth>0.0001</minDepth>
    <mu1>1</mu1>
    <mu2>1</mu2>
    <fdir1>0 0 0</fdir1>
</gazebo>
<gazebo reference="r_foot">
    <collision>
        <surface>
            <bounce>
                <restitution_coefficient>0</restitution_coefficient>
                <threshold>100000</threshold>
            </bounce>
        </surface>
    </collision>
    <maxContacts>4</maxContacts>
    <kp>18000000</kp>
    <kd>100</kd>
    <maxVel>100</maxVel>
    <minDepth>0.0001</minDepth>
    <mu1>1</mu1>
    <mu2>1</mu2>
    <fdir1>
<gazebo>

[traversaro]

As discussed f2f with @DanielePucci , @francesco-romano and @fiorisi , for unblocking this issue the plan is to automatically change the inertia of the small links using the assignedInertia feature of simmechanics-to-urdf script, and by enabling the implicit damping (see #24).

[diegoferigo]

I remember @nunoguedelha some while ago performed some test with implicit damping without much luck

[traversaro]

Yes, but I think that the enabling the implicit spring damping should decrease the amount to which the inertia should be increased.

[nunoguedelha]

Regarding the point "Handling the small inertias." in the checklist, we changed the inertia of some of the smallest robot links. This is to be ported into the URDF generation scripts. The impacted links are listed below, with the respective inertia delta. All the changed link inertia have been set to <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/> as a quick fix:

<link name="r_hip_3">
-      <inertia ixx="3.27938e-05" ixy="2.67901e-07" ixz="-4.41463e-09" iyy="5.69957e-05" iyz="3.32338e-08" izz="3.26449e-05"/>
+      <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/>

<link name="r_ankle_2">
-      <inertia ixx="0.000335682" ixy="-7.53335e-08" ixz="2.93681e-08" iyy="0.00025387" iyz="-4.26218e-07" izz="0.000154833"/>
+      <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/>

<link name="l_hip_3">
-      <inertia ixx="3.27938e-05" ixy="2.67901e-07" ixz="-4.41463e-09" iyy="5.69957e-05" iyz="3.32338e-08" izz="3.26449e-05"/>
+      <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/>

<link name="l_ankle_2">
-      <inertia ixx="0.000335682" ixy="-7.53335e-08" ixz="2.93681e-08" iyy="0.00025387" iyz="-4.26218e-07" izz="0.000154833"/>
+      <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/>

<link name="r_shoulder_1">
-      <inertia ixx="2.24541e-05" ixy="-1.65173e-08" ixz="-3.55626e-06" iyy="2.29582e-05" iyz="-6.13337e-08" izz="1.00116e-05"/>
+      <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/>

<link name="r_shoulder_2">
-      <inertia ixx="0.000158626" ixy="-1.12091e-07" ixz="1.02574e-05" iyy="0.000206526" iyz="1.19785e-06" izz="0.000104956"/>
+      <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/>

<link name="r_shoulder_3">
-      <inertia ixx="0.000142333" ixy="3.60361e-07" ixz="4.94445e-05" iyy="0.000197746" iyz="2.65919e-07" izz="0.000151821"/>
+      <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/>

<link name="r_elbow_1">
-      <inertia ixx="8.13344e-05" ixy="7.36974e-06" ixz="-1.04536e-05" iyy="5.86947e-05" iyz="3.92498e-06" izz="9.68353e-05"/>
+      <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/>

<link name="r_wrist_1">
-      <inertia ixx="5.57722e-06" ixy="-3.10462e-08" ixz="-9.44554e-07" iyy="5.29845e-06" iyz="-1.59685e-07" izz="1.87724e-06"/>
+      <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/>

<link name="r_hand">
-      <inertia ixx="0.000162381" ixy="-2.3536e-06" ixz="3.61526e-05" iyy="0.00043138" iyz="2.07146e-05" izz="0.00030575"/>
+      <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/>

<link name="l_shoulder_1">
-      <inertia ixx="2.24676e-05" ixy="-1.77064e-08" ixz="3.55951e-06" iyy="2.29712e-05" iyz="6.57597e-08" izz="1.00139e-05"/>
+      <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/>

<link name="l_shoulder_2">
-      <inertia ixx="0.000159621" ixy="1.19701e-07" ixz="-1.29687e-05" iyy="0.000222574" iyz="1.20112e-06" izz="0.000120056"/>
+      <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/>

<link name="l_shoulder_3">
-      <inertia ixx="0.00014245" ixy="-2.22441e-07" ixz="-4.95846e-05" iyy="0.000197807" iyz="1.07538e-07" izz="0.000151706"/>
+      <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/>

<link name="l_elbow_1">
-      <inertia ixx="8.13352e-05" ixy="-7.37159e-06" ixz="1.04505e-05" iyy="5.86823e-05" iyz="3.9319e-06" izz="9.68469e-05"/>
+      <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/>

<link name="l_wrist_1">
-      <inertia ixx="5.56774e-06" ixy="3.08362e-08" ixz="9.42153e-07" iyy="5.28838e-06" iyz="-1.58924e-07" izz="1.87579e-06"/>
+      <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/>

<link name="l_hand">
-      <inertia ixx="0.000162491" ixy="2.96453e-06" ixz="-3.68906e-05" iyy="0.000430657" iyz="2.06885e-05" izz="0.000304729"/>
+      <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/>

<link name="neck_1">
-      <inertia ixx="4.0647e-05" ixy="8.09156e-08" ixz="-3.74998e-08" iyy="3.881e-05" iyz="9.39717e-06" izz="4.7535e-05"/>
+      <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/>

<link name="neck_2">
-      <inertia ixx="0.000117943" ixy="-5.25591e-07" ixz="4.68307e-07" iyy="4.04013e-05" iyz="9.81476e-06" izz="0.000105186"/>
+      <inertia ixx="0.01" ixy="0" ixz="0" iyy="0.01" iyz="0" izz="0.01"/>

[nunoguedelha]

@diegoferigo @traversaro Edited the checklist in the initial comment of this issue. I think we can close this issue.

[S-Dafarra]

Both the wrist yaw joints go in hardware fault. Apparently the position control is unstable and it brings the joints toward the limits. I tried to reduce the gains, but it did not work

Apparently my tests weren't that good. Changing the PIDs does the job! Setting 0 for the derivative and the integral part seems to be enough.

[gabrielenava]

iCub models meeting

This afternoon @nunoguedelha, @diegoferigo, @traversaro and I had a meeting about how to use iCub models. These are the outcomes of our meeting:

Definition of the new workflow for models

• All robots .urdf raw models are autogenerated inside icub-models-generator repository. Then, they are copy-pasted inside icub-models repository. A .urdf model for simulation is added to the list of models. Status: already implemented

• All composite models (e.g. icub on a chair, icub seesaw...) are inside icub-gazebo-wholebody repository. Status: almost finished

• icub-gazebo repository still contain an old version of icub .sdf, but it will be deprecated soon. Status: tell people we're going to deprecate icub-gazebo

• With the introduction of WBToolbox 3.0, yarp-wholeBodyInterface is now deprecated.

TODO list:

• simplify collision meshes for the feet

• add estimated joint friction to the model

• gain tuning of the new model for balancing with WBT3.0

• use directly raw models into Gazebo

• overwrite initial joint config. in the composite models

[nunoguedelha]

Edited meeting minutes #52 (comment): converted "TODO list" to a check list.

[fiorisi]

@nunoguedelha can you please update the issue?

[traversaro]

The only issue still open related to this issue is #56 . However, that issue does not prevent to use the models in Gazebo, and so I think we can close this issue.