[simheo]

Follow-up to #27: the Orbbec Gemini 305 replaced the OAK-D SR as the default depth camera, and #27 made the prose generic. This renames the OAK-specific identifiers that were held back so in-flight branches wouldn't need painful rebases. dcam (depth camera) is the generic prefix, as for the raw dcam_* files.

Renames (all git mv, so history follows)

Before	After
grabette_postprocess/oak_slam.py, run_oak_slam()	slam.py, run_slam()
scripts/pipeline/run_oak_slam.py	run_slam.py
scripts/pipeline/convert_episode_to_oak.py	convert_episode_for_slam.py
docker/oak_vslam/	docker/vslam/
check_oak_imu, load_oak_gyro_norm, oak_left_frame_ts	check_dcam_imu, load_dcam_gyro_norm, dcam_left_frame_ts
private helpers / locals (_check_oak_cameras, _load_oak_left_timestamps, _load_oak_imu, oak_dir, oak_ts…)	dcam / slam_dir equivalents
per-episode SLAM input dir oak/, log oak_slam_stdout.txt	slam_input/, slam_stdout.txt
base image pollenrobotics/oak-vslam	pollenrobotics/grabette-vslam

pipeline.py already had its own batch run_slam(), so it now imports the module (from grabette_postprocess import slam, slam.run_slam(...)) instead of the function.

Also in here

• IMU-less log line: a Gemini run printed no IMU gravity ([Errno 2] No such file or directory: …/oak/imu_acc.csv) for every episode. A missing imu_acc.csv is the normal IMU-less case and now logs no IMU (IMU-less camera): …. Other IMU failures keep the real error. Two tests added (tests/test_slam.py).
• Leftover OAK wording in dataset.py, slam.py, app.py, pipeline.py.
• dataset.py: the h264 rationale comment from the monorepo copy, and a needless f-prefix dropped.

Docker image

pollenrobotics/grabette-vslam:2026-07-22 and :latest are the same image as oak-vslam, with the same digest sha256:3fd1c9…. It was re-pushed, not rebuilt. oak-vslam stays as a frozen alias. Both names are public.

Kept on purpose

• legacy oakd_* / oak_mask.png names in episode_files.py and their tests (old recordings on the Hub)
• OAK-D-specific notes, and references to grabette's hardware/oakd.py
• the "oak" camera label inside offline_vslam.cpp
• the attribution in THIRD_PARTY_NOTICES.md; only its two docker/vslam/ paths changed, since the old ones would point at nothing

git grep -n -i -P '(?<![a-z])oak' -- . ':!*.lock' now shows only these.

Behaviour change

New runs write slam_input/ and slam_stdout.txt. Episodes converted earlier still have oak/. Re-running the pipeline converts them again, but visualize_rgbd_trajectory.py --gravity-align on an old processed episode needs a re-convert first.

Verification

• pytest: 70 passed (68 + 2 new)
• --help runs for all 11 scripts (the visualizer with --extra viz)
• pipeline.py imports; slam.run_slam resolves to the package, run_slam to the batch runner
• Local SLAM through the renamed scripts (convert_episode_for_slam.py → run_slam.py) with the default grabette-vslam image: OAK-D episode 144/144 tracked, writes slam_input/, slam_stdout.txt, camera_trajectory.csv. A Gemini episode fails with nlohmann type_error … null; that bug already exists on develop: the published image (built 2026-07-22) predates IMU-less support in offline_vslam.cpp (590a794, 2026-08-18). The Space is unaffected because its Dockerfile recompiles the binary. A rebuilt image is a separate follow-up.
• Test Space: b4fad29 built (FROM grabette-vslam, offline_vslam compiled from docker/vslam/), page shows the branch, API answers. A full conversion of simheo/test_0909_gemini-raw (4 Gemini episodes) logged slam_input/ written and the plain no IMU (IMU-less camera) line, tracked 4/4 at 100% (all GOOD), and pushed the dataset

🤖 Generated with Claude Code