fix: add try-except block for Isaac Lab compatibility (#9)

[zbbright233]

What does this PR do?

Fixes an AttributeError: 'Tensor' object has no attribute 'torch' when running training scripts with recent Isaac Lab updates (Extension version 0.54.4+), where physics buffers now natively return standard PyTorch Tensors.

How is it resolved?

To maintain backward compatibility for users still on older versions of Isaac Lab, I wrapped the data fetching in a try-except block. This ensures the code runs flawlessly on both old and new versions of Isaac Lab.

try:
    # Works for older Isaac Lab versions
    self.forwards = math_utils.quat_apply(self.robot.data.root_link_quat_w.torch, self.robot.data.FORWARD_VEC_B.torch)
except AttributeError:
    # Works for recent Isaac Lab (0.54.4+)
    self.forwards = math_utils.quat_apply(self.robot.data.root_link_quat_w, self.robot.data.FORWARD_VEC_B)

Fixes #9

[greptile-apps]

Greptile Summary

This PR fixes a compatibility issue in _get_observations where root_link_quat_w and FORWARD_VEC_B used to return wrapped buffers requiring a .torch accessor, but now return plain PyTorch Tensors on Isaac Lab 0.54.4+. The implementation (which differs from the try-except shown in the PR description) correctly uses hasattr to detect the API shape.

• Replaces the per-step try-except anti-pattern (originally filed in [Bug] AttributeError: 'Tensor' object has no attribute 'torch' in direct task template with recent Isaac Lab (0.54.4) #9) with a hasattr(quat, "torch") guard that correctly selects the accessor path at call time.
• The API-version check runs every environment step during training; since the Isaac Lab version is fixed for the lifetime of a process, the result of hasattr never changes — caching it once in __init__ would eliminate the repeated attribute lookup at no cost.

Confidence Score: 5/5

Safe to merge — the change is a narrow, well-scoped fix that correctly handles both Isaac Lab API shapes without breaking existing behavior.

The hasattr-based guard correctly replaces the original hard-coded .torch calls and is functionally correct for both old and new Isaac Lab versions. No correctness or data-loss risk is introduced.

No files require special attention.

Important Files Changed

Filename	Overview
source/isaac_lab_tutorial/isaac_lab_tutorial/tasks/direct/isaac_lab_tutorial/isaac_lab_tutorial_env.py	Replaces the original .torch accessor calls with a hasattr-guarded block, correctly supporting both old and new Isaac Lab API shapes; minor inefficiency of re-evaluating hasattr every step.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["_get_observations() called\n(every env step)"] --> B["Fetch root_link_quat_w\nand FORWARD_VEC_B"]
    B --> C{"hasattr(quat, 'torch')?"}
    C -- Yes\nOld Isaac Lab --> D["quat = quat.torch\nfwd_vec = fwd_vec.torch"]
    C -- No\nIsaac Lab 0.54.4+ --> E["Use tensors directly"]
    D --> F["math_utils.quat_apply(quat, fwd_vec)"]
    E --> F
    F --> G["Compute observations\n(dot, cross, forward_speed)"]

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
flowchart TD
    A["_get_observations() called\n(every env step)"] --> B["Fetch root_link_quat_w\nand FORWARD_VEC_B"]
    B --> C{"hasattr(quat, 'torch')?"}
    C -- Yes\nOld Isaac Lab --> D["quat = quat.torch\nfwd_vec = fwd_vec.torch"]
    C -- No\nIsaac Lab 0.54.4+ --> E["Use tensors directly"]
    D --> F["math_utils.quat_apply(quat, fwd_vec)"]
    E --> F
    F --> G["Compute observations\n(dot, cross, forward_speed)"]

Loading

_{Reviews (2): Last reviewed commit: "refactor: optimize compatibility check u..." | Re-trigger Greptile}

[greptile-apps]

Exception thrown on every training step on new Isaac Lab

On Isaac Lab 0.54.4+, the try branch always raises AttributeError (because .torch no longer exists), which is then caught by except. Since _get_observations is called every environment step during training — potentially millions of times — this results in millions of exceptions being raised and swallowed silently. Python exception handling has non-trivial overhead compared to a conditional check, and the pattern also risks masking unrelated AttributeErrors raised inside quat_apply itself.

Prefer detecting the API shape once (e.g., with hasattr) rather than relying on exception control flow in a hot loop.

Suggested change

	try:
	# Works for older Isaac Lab versions where buffers are wrapped
	self.forwards = math_utils.quat_apply(self.robot.data.root_link_quat_w.torch, self.robot.data.FORWARD_VEC_B.torch)
	except AttributeError:
	# Works for recent Isaac Lab (0.54.4+) where buffers natively return standard PyTorch Tensors
	self.forwards = math_utils.quat_apply(self.robot.data.root_link_quat_w, self.robot.data.FORWARD_VEC_B)
	# Support both older Isaac Lab (wrapped buffers with .torch accessor) and
	# newer Isaac Lab 0.54.4+ (native PyTorch Tensors returned directly).
	quat = self.robot.data.root_link_quat_w
	fwd_vec = self.robot.data.FORWARD_VEC_B
	if hasattr(quat, "torch"):
	quat = quat.torch
	fwd_vec = fwd_vec.torch
	self.forwards = math_utils.quat_apply(quat, fwd_vec)

[zbbright233]

Thanks for the great feedback regarding the performance overhead in the hot loop! I have refactored the compatibility logic using hasattr as suggested to keep the execution fast and clean.