Remaining review needs#

This page tracks missing facts and reproduction checks. It distinguishes information an author or hardware inspection must supply from software checks that another contributor can carry out.

Author and hardware review#

Topic

Input needed

Why it matters

Dex3 installation

Deck now establishes the 4-pin, 300 mm JST-GH extension and photographed routing/fitting; still need pinout, power-isolation steps, present fastener condition and linked boot video

Complete the walkthrough without generalizing this robot’s damaged threads or using slide order as an operating procedure

Current robot

Head-pitch witness mark now photographed; still record the complete intended support, harness, hands and waist configuration together

The physical configuration changed between June and September

Marker mounts

Final prints on both hands, third-pad contact, screw fit and target orientation

CAD dimensions alone do not prove the physical mount seats correctly

Torso carrier

Lower M6 reach, measured upper engagement, and complete V7 assembly/board repeatability

V7/R2 is current; partial coupon trials and recorded CAD checks do not qualify the torso reference

Licensing

Documentation license and asset attribution review

A public repository is not itself a blanket reuse license

Connect demonstrations to their source runs#

The supplied physical stacking video now shows pickup, placement, release and hand withdrawal. Match it to the retained bag/run ID and cleanup outcome; the footage alone does not establish software ownership handback. The standing calibration video now shows the wrist markers and physical pose sequence; attach its exact session ID as well. The simulated retention clip also needs its candidate ID and generating report. For each, retain the run ID, date, source version and outcome. If a clip stops before return or cleanup, say so in the caption.

The media catalog identifies source figures and publication copies. The retention and stacking-plan demos are rendered/simulated; keep physical success claims tied to physical runs.

Evidence and reproducibility gaps#

  • Review the operator runbook against the exact intended deployment revision. Test installation in a clean workspace; the setup chapter identifies remaining lab paths, message dependencies and PC2 prerequisites.

  • The newest manually preclosed calibration workflow needs regenerated cores and physical validation before it can be an operating procedure. Old closed-route geometry is not reusable as-is; this is a capability limit, not a documentation instruction to run an experiment.

  • Independently annotate physical placement/stack success before reporting a success rate from the August 21 or August 25 completion records.

  • Reproduce the MuJoCo setup in a clean workspace and reconcile its older model-shape diagnostic. The pretrained OpenHomie policy is linked in the chapter; sampled demo frames were reviewed, without a quantitative stability test.

  • The public dataset viewer loads the five demo episodes and both video streams. Its clean dependency installation still needs a separate reproduction; the walkthrough’s additional local viewer changes are not all public.

  • Assembly reproduction needs the generated assets and qualification pools listed by the source. Published scripts and selected reports are not a full historical experiment archive.

Handoff and presentation#

Preserve private experimental records separately, retain access to the Drive datasets and release videos, and resolve the license/attribution review before the lab handoff. Maintenance describes the required repository, storage and ownership updates.

Desktop layout remains the current presentation target. Revisit mobile navigation, wide tables and diagram interaction during the final presentation pass after the chapter content is settled.