Torque-only control mode

[xmatousm]

In our lab we are controlling the robots using torque-only mode (feedback controlling loop is computed in the ROS). To be able to do this, we set the stifness in sunrise workbench java part to zero. There are issues that we had to solve:

• robot still needs to send joint position also - not used as stifness is zero, but must be not so far from actual one due to built in safety checks
• when internal gravity compensation is not precise enough (which seems it isn't) the startup must be carefully solved, as ROS nodes startup takes some time and (at least in our architecture) the HW interface starts (and robot releses it brakes) a small amout of time before the full (rather complex) controller.

We have solved these issues by implementing the 'torque_only' parameter and client mode to system_interface, creating the new torque_only_command interface and by implementing simple P-control loop inside it that is active before the first real control torques are obtained.

Is this interesting for main-stream to create a merge request from our side?

[mhubii]

Sorry for the late reply @xmatousm, please feel free to tag!

Yes, happy to have a look at this. Is your addition compatible with the existing code, or do you anticipate incompatabilities?

[xmatousm]

No problem, I'm working on multiple projects as well...

My change seems to be compatible with existing code that uses the lbr, as it adds a new 'mode'. Yet some changes must be done to one or two function signatures of the internals of hw interface itself (we need to pass and information that the mode is torque-only, and the enum of FRI modes has no value for this.

I will prepare the PR, so that you can have a look on this. Though it would be maybe difficult for me to do (and test) this in rolling, as we have all the stuff in humble...