Run our own controllers in the controller manager

[Steven Gray]

Are we allowed to run our own controllers for the arm - i.e. stop the running ariac/arm controller and add our own to the controller manager? I realize the controller manager is part of the Gazebo instance.

[Deanna Hood]

Hi @stgray, just letting you know that we've seen your question and will get back to you with a response.

[Deanna Hood]

Could you clarify your intent? You'd be aware that there is an effort position controller by default. Did you have an interest in a different type of controller, different parameters, or something else?

[Steven Gray]

I was thinking of implementing different controllers. I'm not sure how well an impedance controller would work without a wrist FT sensor, but I might dust off ee_cart_imped_control and use the reported joint efforts to estimate the end effector wrench. I could also make do with a gravity compensation controller and a Jacobian pseudo-inverse controller.

[Deanna Hood]

Thanks for your patience.

I think you're aware, but for anyone else reading: the only control interface to the robot is via gazebo_ros_control which uses ros_control, and is launched as part of the ARIAC server (in the Gazebo instance, as you mentioned). There's an EffortJointInterface as the hardware interface to the robot that cannot be changed.

ros_control as is does not support the controllers that you propose (see for example this issue). For you to use a custom controller you would need to be able to load a version of ros_control that supports it as part of the ARIAC server, and that level of configuration is not available to teams. There are no plans to add this functionality, given that the focus of the competition is on higher-level agility strategies.

[Steven Gray]

No worries, I had a feeling that would be the case.

I'll probably jump onto #126; I have noticed a number of times where the final joint angles of multiple joints are ~0.1 radians off from the desired at the end of trajectory execution.

[Deanna Hood]

the joints can be a bit slow to converge; it's something we've attempted to address in the latest releases. I'll close this particular issue for now