Fixes tensor construction warning in events.py

[louislelay]

Description

This PR removes a UserWarning from PyTorch about using torch.tensor() on an existing tensor in events.py. It replaces torch.tensor(actuator.joint_indices, device=asset.device) with .to(device) to avoid unnecessary copies.

Warning mentionned:

/home/spring/IsaacLab/source/isaaclab/isaaclab/envs/mdp/events.py:542: 
UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor). actuator_joint_indices = torch.tensor(actuator.joint_indices, device=asset.device)

Type of change

• Bug fix (non-breaking change which fixes an issue)

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

[ooctipus]

@louislelay

So actuator.joint_indices can only be torch.tensor or slice? if that is true maybe add if isinstance(slice), elif isinstance(torch.Tensor), else raise error may be a bit more readable?

On a second note, do you think check the type during call time can be too expensive (I know you didn't write the original function, so no pressure to edit), (◎ ◎)ゞ why couldn't we do the check and construct indices during init construction instead?

Would love to know your thought on it. : ))

[Mayankm96]

So actuator.joint_indices can only be torch.tensor or slice? if that is true maybe add if isinstance(slice), elif isinstance(torch.Tensor), else raise error may be a bit more readable?

Yes, i think it makes sense to become more strong-type if the type is known already.

Why couldn't we do the check and construct indices during init construction instead?

We do the checks for findding joint indices when resolving scene entities. Do you mean the check in the event term itself?

[louislelay]

Hey @ooctipus and @Mayankm96, thanks for the feedback!

I’ve updated the code to make the checks type-explicit and raise on unexpected types. I’ll run tests once my current training job finishes and let you know when it’s ready for review again. Edit: it's good to me :)

On the second point: I agree with Octi, moving the joint index resolution to __init__ makes sense since it stays stable during training and avoids extra work each step.

[louislelay]

I tested it and accepted your changes @Mayankm96, aside from the init vs call thingy the rest is good to me!

[ooctipus]

@louislelay

- global_indices = actuator_indices = torch.tensor(self.asset_cfg.joint_ids, device=self.asset.device)
+ global_indices = torch.tensor(self.asset_cfg.joint_ids, device=self.asset.device)

ahh we are now missing actuator_indices and can cause error because it is not defined
I will submit an PR to fix this