LAB / 029

Embodied AI practice (14): does landing in the bin mean the task is complete?

Eight MuJoCo trajectories, three passive readouts and recorded-state replay separate historical placement, current evidence and qualified completion.

Completed ·

EXPERIMENT RESULT

GitHub ↗

01 / QUESTION

Eight MuJoCo trajectories, three passive readouts and recorded-state replay separate historical placement, current evidence and qualified completion.

METHOD

Eight fixed MuJoCo rollouts with three passive readouts each; 48,000 raw rows, 4,800 states and a three-scenario replay.

RESULT

Lowering-drop passes physical placement but remains ABORTED; post-release push retains historical completion but becomes INVALID.

OBSERVATIONS

Silence and stale replay expire at 8.042s; a single false bad packet also revokes validity and requires a new stable window.

CONCLUSION

Completion needs execution prerequisites and continuing evidence; store history, current physical state and controller termination separately.

LIMITATIONS

Release is command intent only. Observers do not recover invalidated tasks; noise and hardware safety remain open, as do full VL01 and M4/M5.

REPRODUCE

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

KEEP EXPLORING

Related content

  1. Engineering cases

    Embodied AI practice (14): does landing in the bin mean the task is complete?

    Eight MuJoCo trajectories, three passive readouts and recorded-state replay separate historical placement, current evidence and qualified completion.

    13 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.

    Experiment
  3. 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.

    13 MINPublished
  4. 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.

    11 MINPublished

← Back to index