Embodied AI practice (13): where should recovery resume after interrupted lowering?

01 / QUESTION

18 MuJoCo rollouts and synchronized replay connect phase contracts to recovery budgets, separating false holds, physical placement and controller abort.

METHOD

Six fixed conditions × three policies, eighteen twelve-second MuJoCo rollouts. HOLD preserves phase; terminal exit always holds.

RESULT

Twelve rollouts meet physical acceptance, including two with aborted controllers. Phase-aware recovery avoids clean false holds and resumes both lowering gaps.

OBSERVATIONS

Late-gap HOLD occurs at 6.142s and resumes at 6.542s; copied transfer height rules falsely hold clean lowering at 6.002s.

CONCLUSION

Model phase, execution state and physical outcome separately; reaching the target need not mean recovery succeeded.

LIMITATIONS

Fixed simulation; permanent observation loss does not complete placement. Accidental placement after opening is not recovery success. No regrasp or hardware safety claim.

REPRODUCE

Pinned code and raw records are in the repository; the article provides run, audit and replay commands.

1. Engineering cases

   Embodied AI practice (13): where should recovery resume after interrupted lowering?

   18 MuJoCo rollouts and synchronized replay connect phase contracts to recovery budgets, separating false holds, physical placement and controller abort.

   2026-10-0413 MINPublished
2. PROJECT / 003

   Hohoo Embodied Experiments

   Two MuJoCo teaching scenes—pick-and-place and planar force control—record contracts, tuning, clearance and delayed-feedback outcomes, including failures.

   2026-09-28Experiment
3. Engineering cases

   Embodied AI practice (12): what should the gripper do after recovery is exhausted?

   12 MuJoCo rollouts and synchronized three-view replay compare hold, open and open-retreat, separating terminal decisions from physical consequences.

   2026-10-0312 MINPublished
4. Engineering cases

   Embodied AI Practice VII: Why a Transfer Rule Stops Normal Lowering

   21 MuJoCo rollouts separate stage-specific grasp checks, stop actions and final placement. A false alarm and lingering finger contact reveal two different failure modes.

   2026-10-0113 MINPublished
5. Engineering cases

   Embodied AI practice (11): placement completed, but is it still valid?

   Seven MuJoCo rollouts and a paired replay separate historical completion from current validity, including false alarms and expiry boundaries.

   2026-10-0311 MINPublished