Robot model moving itself

[ChristopherAbram]

I tried this repository for a research purpose. After some time playing with it, I noticed that mobile base is moving, even if /cmd_vel topic is empty. The base is also moving very slowly right after enabling physics engines in gazebo.
I have done some review of a urdf model and all launch files, and I've noticed that it might be caused by disabling friction for wheels. I tried to change it, and if I set it in this way (in file mir_100.gazebo.xacro)

<xacro:macro name="set_all_wheel_frictions" params="prefix">
    <xacro:set_wheel_friction link="${prefix}left_wheel_link" friction="0.4"/>
    <xacro:set_wheel_friction link="${prefix}right_wheel_link" friction="0.4"/>
    <xacro:set_wheel_friction link="${prefix}fl_caster_wheel_link" friction="0.4"/>
    <xacro:set_wheel_friction link="${prefix}fr_caster_wheel_link" friction="0.4"/>
    <xacro:set_wheel_friction link="${prefix}bl_caster_wheel_link" friction="0.4"/>
    <xacro:set_wheel_friction link="${prefix}br_caster_wheel_link" friction="0.4"/>
</xacro:macro>

and small modification in mir_100_v1.urdf.xacro

<!-- disabled, because it doesn't make a difference in Gazebo: -->
<xacro:set_all_wheel_frictions prefix="${prefix}"/>

then the model was still.
I haven't calculated friction coefficients, I just have used arbitrary chosen values (for which simulator seems to be stable).

[mintar]

Thanks for your contribution. I've made some experiments myself (see below). The short summary is that I cannot see a change in behavior after applying your suggestions. Also note that you reduced the friction; by default it's 1.0, you reduced it to 0.4.

I've also let the simulation run for 2 minutes and 18 seconds (simulation time). Nothing special about that value, it's just the time after which I killed the first simulation. I've included the output of the /base_pose_ground_truth topic below; the robot moved between 0.4 and 6 mm forward during that time. I don't think that will be noticeable.

no modification (current kinetic branch, 54be543)

mu1 = mu2 = 1.0
kp = 2,147,483,647.0
kd = 1

rostopic echo /base_pose_ground_truth
pose: 
  pose: 
    position: 
      x: 0.00641750607671
      y: -0.00512987328118
      z: -1.1398489202e-05

only enable set_all_wheel_frictions with current URDF values (friction = 200 for wheels, 1 for casters)

mu1 = mu2 = 200 (wheels) / 1 (casters)
kp = 10,000,000
kd = 1

rostopic echo /base_pose_ground_truth
pose: 
  pose: 
    position: 
      x: 0.000424151391665
      y: -2.32785455031e-06
      z: -5.94617817326e-06

However: If you use rqt_robot_steering to make the robot turn in place and then
stop abruptly, the casters spin around funnily. That doesn't happen in the other two cases.

your proposal (friction = 0.4 for wheels + casters)

mu1 = mu2 = 0.4
kp = 10,000,000
kd = 1

rostopic echo /base_pose_ground_truth
pose: 
  pose: 
    position: 
      x: 0.00200195846341
      y: -7.47403570384e-06
      z: -7.08789661061e-06

Same behavior as current kinetic branch.

---

Since I cannot reproduce the bug, I don't think I'm going to merge this, sorry. :)

[mbed92]

Hey. Pretty old issue, but I also faced that problem recently. In my case, it was visible after ~30min and the robot rotated for ~4 degrees in the Z-axis. In my case, I changed the robot's model to fit my purposes, but the simulation part was left unchanged. Changing friction slowed down the rotation, but not stopped.

[mintar]

First of, I should mention that we did re-enable set_all_wheel_frictions in #19, after we could reproduce the bug in some other Gazebo environments.

In response to you comment: 4 degrees in 30 minutes don't sound much.

However, if you find any parameters that improve the situation, feel free to open a pull request!