[Test] Stabilize RTX MDL shader-warmup flake in rendering correctness tests

source/isaaclab_tasks/changelog.d/jichuanh-stabilize-rendering-mdl-warmup.rst

@@ -0,0 +1,13 @@
+Fixed
+^^^^^
+
+* Stabilized ``test_camera_renders_not_empty`` in
+  ``test_shadow_hand_vision_presets.py`` by polling the camera output until
+  all data-type tensors are non-zero, with a 60-step cap, instead of a
+  single ``env.step()``. The test intermittently failed with "Camera output
+  is all zeros or all inf" for ``simple_shading_*_mdl`` and
+  ``simple_shading_constant_diffuse`` variants on cold-cache CI runners
+  because the GPU returned a still-zero framebuffer before the MDL material
+  finished compiling. The three goldenfile-comparing helpers in
+  ``rendering_test_utils.py`` already use ``flaky(max_runs=3)`` and are left
+  untouched.

source/isaaclab_tasks/test/test_shadow_hand_vision_presets.py

@@ -398,7 +398,21 @@ def render_correctness_env(request, shadow_hand_vision_presets):
     env = ShadowHandVisionEnv(cfg)
     env.reset()
     actions = torch.zeros(cfg.scene.num_envs, env.action_space.shape[-1], device=env.device)
-    env.step(actions)
+    # Step until all camera outputs are non-zero (RTX MDL shaders compile lazily).
+    # ``simple_shading_*_mdl`` presets can take 10–30 frames to produce non-zero pixels
+    # on cold-cache CI runners; poll up to ``_MAX_WARMUP_STEPS`` and exit early once ready.
+    _MAX_WARMUP_STEPS = 60
+    for _ in range(_MAX_WARMUP_STEPS):
+        env.step(actions)
+        outputs_ready = True
+        for output in env._tiled_camera.data.output.values():
+            tensor = output.torch
+            finite = torch.where(torch.isinf(tensor), torch.zeros_like(tensor), tensor)
+            if finite.max() <= 0.2:
+                outputs_ready = False
+                break
+        if outputs_ready:
+            break
     yield renderer_preset, camera_preset, physics, env
     env.close()