feat: add clip range of JointAction

[fan-ziqi]

Description

Add action clip to all mdp/actions

Type of change

• New feature (non-breaking change which adds functionality)
• This change requires a documentation update

Checklist

• I have run the pre-commit checks with ./isaaclab.sh --format
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• I have updated the changelog and the corresponding version in the extension's config/extension.toml file
• I have added my name to the CONTRIBUTORS.md or my name already exists there

[jtigue-bdai]

Currently limits are handled by the actuators. Primarily this is done on the effort.

[fan-ziqi]

So how can I clip actions in isaaclab like in legged_gym?

[jtigue-bdai]

Hmm I guess if you are using joint position commands the actuator limits won't really handle that. Actuator limits really only do effort.

I will take another look at the current api and get back to you.

[fan-ziqi]

Thanks for reply!
Yes, I use Δjoint_pos as actions

[kellyguo11]

The discussion in this PR may be relevant - #984. This seems like something that would make more sense for the wrappers.

[fan-ziqi]

Which method is better? I don't know if this should be an env feature or an action feature

[jtigue-bdai]

Yeah #984 makes the clip for all actions where this PR will makes the clip configurable for each action separately. To me it makes sense to have the action handle the clipping rather than the environment.

[jtigue-bdai]

I guess this one only handles the joint_actions.

[fan-ziqi]

For me, this allows more flexibility in configuring each joint individually.

So is this PR necessary?
Should we adopt this or #984?

[kellyguo11]

That's a good point, this could be useful for configuring the joints separately, but only applicable for joint actions. Perhaps there's value in adopting both approaches.

[jtigue-bdai]

I could see both and also adding the clip configuration at the base ActionTerm/Cfg. It would just require that the actual processing be added to all provided actions.

[jtigue-bdai]

Hey @fan-ziqi would you be willing to add the clipping cfg to the top level ActionTermCfg, the clip checking in the top level ActionTerm.init , and then add the clipping execution to all the provided actions in the envs/mdp/actions directory?

[fan-ziqi]

I have added clip to ActionTermCfg and added clip action to all mdp/actions

But I have a question, is clip: dict[str, tuple] necessary?

@configclass
class ActionsCfg:
    """Action specifications for the MDP."""

    joint_pos = mdp.JointPositionActionCfg(
        asset_name="robot", joint_names=[".*"], scale=0.5, use_default_offset=True, clip={".*": (-100.0, 100.0)}
    )

or clip: tuple[float, float] is enough

@configclass
class ActionsCfg:
    """Action specifications for the MDP."""

    joint_pos = mdp.JointPositionActionCfg(
        asset_name="robot", joint_names=[".*"], scale=0.5, use_default_offset=True, clip=(-100.0, 100.0)
    )

[jtigue-bdai]

I think being able to do clip=dict[str,tuple] is necessary to be able to set clips on a per joint basis.

[kellyguo11]

Thanks a lot for the feature! would it be possible to add some unit tests for it as well?

[kellyguo11]

is it possible to resolve this once at initialization and convert to a torch tensor to avoid more overhead with the loop?