unable to use motor without PID setting

[christianrauch]

Describe the bug
I am trying to use a motor actuator

<actuator>
    <motor class="x2" name="rotor/1" site="thrust1" gear="0 0 1 0 0 -.0201"/>
</actuator>

But this fails with something like:

Velocity command interface for the joint : rotor/1 is not supported with motor or custom actuator without defining the PIDs

It is unclear which PID controllers this is referring to, and the MujocoSystemInterface cannot assume any controllers loaded in the ros2_control_node.

To Reproduce
Steps to reproduce the behavior:

1. load the Skydio X2 with support "site" transmission type #154

Expected behavior
As hardware_interface::SystemInterface, MujocoSystemInterface should only provide control and state interfaces for controllers, and not force users to configure controllers inside the SystemInterface.

Screenshots
If applicable, add screenshots to help explain your problem.

Environment (please complete the following information):

• OS: [e.g. Ubuntu]
• Version [e.g. Humble]
• Anything that may be unusual about your environment

Additional context
Add any other context about the problem here, especially include any modifications to ros2_control that relate to this issue.

[saikishor]

@christianrauch As the error specifies, the system interface doesn't allow you to have a velocity command interface for your ros2_control configuration inside your URDF if the PIDs are not defined, you are fine to use the interface effort/torque for this particular usecase, but not of position and velocity.

[christianrauch]

I understand that MujocoSystemInterface wants the user to define PID gains. But I don't understand the reason behind this.

If I implement a SystemInterface to control a motor, e.g. via PWM, then I would implement a controller, e.g. to control the orientation of a skid-steered robot, via the ControllerInterface or ChainableControllerInterface. I don't see the rationale to use two different control frameworks. Additionally, why is this only for the position and velocity interfaces?

[saikishor]

I understand that `MujocoSystemInterface wants the user to define PID gains. But I don't understand the reason behind this.

If I implement a SystemInterface to control a motor, e.g. via PWM, then I would implement a controller, e.g. to control the orientation of a skid-steered robot, via the ControllerInterface or ChainableControllerInterface. I don't see the rationale to use two different control frameworks. Additionally, why is this only for the position and velocity interfaces?

If you want to use the position or velocity interfaces, you need to have position or velocity actuators and for the generic motor, you can only use the effort interface (or) use the position and velocity with the corresponding PIDs. This is how we can enable all 3 modes in the MuJoCo, there is no other easy way right?

[christianrauch]

This is how we can enable all 3 modes in the MuJoCo, there is no other easy way right?

I don't understand. Sorry.

If I have a real motor whose speed is set open-loop via PWM, I can implement a SystemInterface or ActuatorInterface to write this PWM value to the motor, given e.g. a relative range [0, 1] that I map to a duty cycle. I don't need a PID controller to set the PWM value, and likely, I don't have an encoder on the motor to read the speed back anyway.

If I now want to replicate this interface in MuJoCo, why do I need PID gains in MuJoCo? Why can the MujocoSystemInterface not just forward the raw commands?

I actually have my own SystemInterface implementation using MuJoCo in fewer than 500 lines that does exactly this, but I wanted to switch to the more generic implementation in this very repo. Frankly, I am new to this, and I might have overlooked something. But from my baseline implementation, I know that I can just forward raw values to MuJoCo without the need to implement a PID inside the SystemInterface.

[saikishor]

@christianrauch just remove the interfaces velocity and position from those respective joints from the URDF ros2_comtrol command interface tags and then you should be able to command the motors in effort. That's the most transparent way to have a controller in velocity and then that sets effort interface commands

[christianrauch]

Isn't velocity semantically the closest interface when I control a motor speed via PWM? My understanding is that effort is some kind of torque or moment. But semantically, I have no way to set this to the motors I am commanding.

In fact, these motors are "controlled" by an ESC (Electronic Speed Controller), which takes PWM as input. The duty cycle of the PWM signal is translated to the frequency of the three-phase sinusoidal signal that powers the motors. That is why, semantically, only the velocity interface makes sense.