fix: enhance startup for perception services on first run

[KE7]

• Replace wget with urllib streaming download in weights.py; logs a progress bar every 5% with speed and ETA. Downloads to a .part temp file and renames atomically on completion, preventing the corrupted weights that caused the original crash.
• Replace the fixed 180s retry timeout in _is_ros2_stack_ready with an activity-based stale timeout (default 120s) that resets whenever new ROS2 services/topics/actions appear. Weight downloads no longer cause false timeouts regardless of how long they take.

Purpose

Fix startup failures when running RAI with AMD GPUs (gfx1151 / Strix Halo) where GroundingDINO weight downloads caused timeouts and GPU errors were silently swallowed.

Proposed Changes

• detection_service.py: Replace exc_info=True (unsupported by rclpy logger, raises TypeError) with traceback.format_exc() so GPU errors are actually logged instead of swallowed.
  • weights.py: Replace subprocess/wget with urllib.request streaming download. Writes to a .part temp file and renames atomically on completion, preventing corrupted weight files from interrupted downloads. Logs a progress bar every 5% with speed and ETA.
  • o3de_bridge.py: Replace the fixed 360-retry (~180s) loop in _is_ros2_stack_ready with an activity-based stale timeout that resets whenever new ROS2 interfaces appear. Eliminates false timeouts during first-run weight downloads regardless of how long they take.
  • gripping_points_tools.py: Increase timeout_sec default from 10s to 60s to accommodate GPU inference startup time.

Issues

N/A

Testing

Tested on AMD Radeon 8060S (gfx1151 / Strix Halo) with ROCm 7.0. GroundingDINO detection service initializes successfully with GPU acceleration confirmed.

[maciejmajek]

Thanks! Tested the rai_perception part. Worked on the following cases:

• Standard uniterrupted download
• Various interrupts at various download steps
• Corrupting a file after download - this case throws an error without redownloading

Seems like there are some tests which need an update.

Please let me know if you will take care of that, I can step in.

============================================================================== short test summary info ==============================================================================
FAILED tests/rai_perception/agents/test_base_vision_agent.py::TestVisionWeightsDownload::test_download_weights_success - Exception: Could not download weights: HTTP Error 404: Not Found
FAILED tests/rai_perception/agents/test_base_vision_agent.py::TestLoadModelWithErrorHandling::test_load_model_corrupted_weights - Exception: Could not download weights: HTTP Error 404: Not Found
FAILED tests/rai_perception/services/test_weights.py::TestDownloadWeights::test_download_weights_success - AttributeError: module 'rai_perception.services.weights' has no attribute 'subprocess'
FAILED tests/rai_perception/services/test_weights.py::TestDownloadWeights::test_download_weights_failure - AttributeError: module 'rai_perception.services.weights' has no attribute 'subprocess'
FAILED tests/rai_perception/services/test_weights.py::TestDownloadWeights::test_download_weights_file_too_small - AttributeError: module 'rai_perception.services.weights' has no attribute 'subprocess'
FAILED tests/rai_perception/services/test_weights.py::TestLoadModelWithErrorHandling::test_load_model_corrupted_weights - AttributeError: module 'rai_perception.services.weights' has no attribute 'subprocess'
FAILED tests/rai_perception/tools/test_pcl_detection_tools.py::test_get_object_gripping_points_tool_auto_declaration - AssertionError: assert 60.0 == 10.0
============================================ 7 failed, 771 passed, 4 skipped, 1 deselected, 11 xfailed, 22 warnings in 110.91s (0:01:50) ============================================

[KE7]

@maciejmajek It would be great if you could help me with the tests. Thank you! I don't have rai running e2e yet on my system as I'm battling dependency and GPU issues

[maciejmajek]

Sure @KE7. What seems to be the problem? I should be able to help you

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 73.11%. Comparing base (59d0e89) to head (a3ff4e4).
⚠️ Report is 1 commits behind head on main.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #769   +/-   ##
=======================================
  Coverage   73.11%   73.11%           
=======================================
  Files          82       82           
  Lines        3571     3571           
=======================================
  Hits         2611     2611           
  Misses        960      960           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[KE7]

@maciejmajek I think I have the tests working now. Any pointers on this ROS failure?

[maciejmajek]

@KE7 No idea, this was a random crash we see from time to time on humble.

[maciejmajek]

LGTM, Thanks!