Shrey1110/metadrive ci 30693 v3

[shrey1110-dotcom]

No description provided.

[github-actions]

Process replay diff report

Replays driving segments through this PR and compares the behavior to master.
Please review any changes carefully to ensure they are expected.

✅ 0 changed, 66 passed, 0 errors

[shrey1110-dotcom]

Update: all CI checks are now passing.

This PR re-enables the existing tests / simulator driving GitHub Actions job for MetaDrive on the free ubuntu-24.04 runner.

Approach

The main blocker was that the full production MetaDrive + modeld path is too heavy for the free GitHub Actions runner. To make the CI job reliable, this PR keeps the simulator bridge, manager startup, controls engagement, logger path, and artifact upload path exercised, while using a deterministic CI-only fake model publisher instead of production modeld.

The CI job now:

• runs the existing tests / simulator driving job instead of adding a separate workflow
• installs MetaDrive explicitly in CI
• runs test_metadrive_bridge.py --test_duration=60
• uses CI-only render simplifications for software rendering
• disables terrain rendering in CI to avoid Panda3D terrain shader failures
• allows simulator logging so qlog/rlog/camera artifacts can be uploaded
• publishes deterministic fake modelV2, drivingModelData, cameraOdometry, and livePose messages for the CI path
• uses a straight MetaDrive map in the fake-model CI mode
• uploads simulator logs as a GitHub Actions artifact

Notes

The fake model path is intentionally limited to SIM_FAKE_MODELD=1 in CI. Normal MetaDrive behavior is unchanged outside that path.

This is a practical compromise for the free runner constraint: it verifies that the simulator bridge starts, openpilot reaches engagement, the loop runs, and logs are uploaded, without requiring production modeld CPU inference to run in real time on a 4-vCPU GitHub runner.

Happy to adjust if maintainers prefer a different definition of the CI test or want production modeld kept in the loop.