Humble 4.4 update broke joints

[mr337]

Humble 4.4 update broke joints

[christophfroehlich]

There were quite a lot of changes to 0.4.4, I don't have a glue now where this could come from. Could you try to compile it from source, and use git bisect to find the commit introducing your issue?

[mr337]

@christophfroehlich Here is an example using the above meshes in the original post of this issue https://github.com/mr337/control_mre Instructions and such on how to produce in ROS2 Humble.

[christophfroehlich]

OK, your model sometimes also explodes with the version before the commit, after it has already been standing stable on the ground. By increasing physics/solver/iterations to 500 make the simulation stable, even with the latest version of the HEAD.

This is an indication that your model or the physics solver is not well configured.

A good starting point is always to have a look at the inertia, masses, and contact points (visualize them from the view menu).

• Having a look at the inertias: None has a infeasible high magnitude, but it does not really coincide with your meshes. Pay also attention that every link being in the kinematic chain with non-fixed joints has reasonable high mass+inertia. (setting inertia of any link to zero -> your model explodes).
• The contact points are flickering. This is a well-documented issue of gazebo classic ODE solver, but there are some solver options to play around with.

I hope I could help with this. I'm pretty sure now that this is not a problem with gazebo_ros2_control, the new version just reveals the problem immediately. I close this issue now, feel free to reopen if you don't agree.

[mr337]

No problem, after working git bisect what appears to be the offending commit is d18a887#diff-e14963d1c2ba3a31138bda99525946c355191fe361859b5fb7247fddad6c3a2a

I believe some of this control code is a little over my head. What is odd is I compared the full ROS log line for line with version 0.4.2 and 0.4.4 expecting some kind of error or something crashing/segfaulting. I did not get any warning as defined here https://github.com/ros-controls/gazebo_ros2_control/pull/177/files#diff-e14963d1c2ba3a31138bda99525946c355191fe361859b5fb7247fddad6c3a2aR138

Looking at the code I don't see much difference since the larger if else blocks seem to cover Position controllers just like I am using. The only difference is they have an additional true. Still splunking around in the codebase.

[christophfroehlich]

I don't see what #208 should change with your setup. Could you provide a full example of your robot including the meshes? I'm curious..

[mr337]

That is what I keep thinking 😆

Let me provide a Minimal Reproducible Example of the robot

[christophfroehlich]

Sorry, I haven't had time earlier to dig into this.

The problem is not a broken joint-state, but your gazebo model explodes after touching the ground.

The reason why this didn't happen before is that now as a fallback every joint velocity is set to zero (this happens always at the first write until the controllers get activated)

gazebo_ros2_control/gazebo_ros2_control/src/gazebo_system.cpp

Lines 604 to 607 in 3c8f5bb

	} else if (this->dataPtr->is_joint_actuated_[j]) {
	// Fallback case is a velocity command of zero (only for actuated joints)
	this->dataPtr->sim_joints_[j]->SetVelocity(0, 0.0);
	}

Without forcing it to zero they are "not actuated"/compliant and the impact is solved stable.

I have to try if this is a problem of your specific robot (falling down, four wheels might not get stable contact points..) or happens also with others.

[mr337]

Just wanted to add that our issue was related to the mass of the wheels. We also did some further work on masses and inertias to get a model that is happy.

Thanks again for pointing us in the right direction @christophfroehlich

[christophfroehlich]

you're very welcome, glad you fixed it!