Actuator Type and PD Control Implementation in Locomotion Settings

[hdadong]

In the locomotion settings, most robots defined in the XML files use the position actuator type. From my understanding, this setup is intended to mimic a PD controller via position control.

However, I noticed that the kd term is implemented using joint damping, as seen here:
https://github.com/google-deepmind/mujoco_playground/blob/main/mujoco_playground/_src/locomotion/t1/xmls/t1_mjx_feetonly.xml#L102

This raises a few questions:

1. Why not use the kv parameter in actuator/position for implementing the derivative (velocity) term of the PD controller? What is the fundamental difference between using joint damping versus kv in the actuator definition?
2. Why choose position actuators instead of motor actuators? Are there specific benefits in terms of stability, controllability, or compatibility with your framework that motivated this design choice?

[vishcomestrue]

Based on my understanding,

1. The fundamental difference between joint damping and kv parameters are that joint damping is a passive force, which means that even if there is no control signal, there is a torque $\tau_{d}\ \alpha\ v$. But the kv term comes to play only when there is a control signal on that actuator. (Refer to the damping term in https://mujoco.readthedocs.io/en/stable/XMLreference.html#body-joint)
2. It could be because of the fact that with position actuator, MuJoCo handles the PD computations and gives us the options to choose k_p, k_d. But motor actuator is more of like a direct torque and would be amplified based on the gear parameter. (Reference: https://mujoco.readthedocs.io/en/stable/XMLreference.html#actuator)

I am not related with the creators of this repository, answers are based on my own understanding. Hope this helps!!