Motor Observation Range and Actuator Model in Sim2Real Setup

[lde-RL]

Thank you for sharing such great content!
I’m curious about how you addressed the reality gap in actuators.
It looks like you're using ImplicitActuatorCfg for the motors—would you consider it a reliable actuator model?
Also, could you share how you define the clamping range for the observations?
For example, is the motor angle clamped to a specific range such as [-2π, 2π]?
I’m currently using Dynamixel MX-64 motors.
Thanks again for the great content!

[jazonshou]

Hey there - thanks for the feedback! I'm checking in with some folks who know more, and I'll get back to you as soon as I have more to share.

[jazonshou]

Hello, thanks for your patience. Regarding your question about clamping, the motor angle's range is clamped between -π and π.

As for your question about ImplicitActuatorCfg, we actually use DCMotorCfg. Unfortunately there isn't really a straightforward way of answering whether or not it is reliable. If you would like, you could give us some more information about your setup and we might be able to do some very vague estimates. But ultimately, we recommend testing things out for your specific context and checking if it's reliable enough for your purpose. Sorry for not having a more straightforward answer :).

Leaving this open in case other folks want to chip in.

[geiko246]

Angles are generally clamped to -π and π through their conversion from quaternions in the native Isaac Lab math_utils. I haven't found that this matters too much.

To expand a bit, the DCMotorCfg is used for the throttle motors to help model any relevant nonlinearities and motor saturation. ImplicitMotorCfg is still used for the steering motors:

WheeledLab/source/wheeledlab_assets/wheeledlab_assets/hound.py

Lines 4 to 22 in 09594e5

	HOUND_ACTUATOR_CFG = {
	"steering_joints": ImplicitActuatorCfg(
	joint_names_expr= ["front_left_wheel_steer", "front_right_wheel_steer"],
	velocity_limit=10.0,
	effort_limit=3.2,
	stiffness=100.0,
	damping=10.,
	friction=0.,
	),
	"throttle_joints": DCMotorCfg(
	joint_names_expr=[".*throttle"],
	saturation_effort=1.05,
	effort_limit=0.25,
	velocity_limit=450.,
	stiffness=0,
	damping=1000.,
	friction=0.0,
	),
	}

We haven't done any rigorous experiments to empirically validate these choices since our downstream results could be achieved just using the manufacturer specification sheet data. This is our motor.