E5 / Revalidating recovery

01 / QUESTION

Does delivery recovery authorize the old action? Compare fresh evidence, replay and remaining-path planning.

METHOD

Six conditions × three policies, 18 fixed rollouts. Control 2 ms, capture 20 ms; recovery requires 100 ms of fresh grasp evidence, with a 600 ms hold deadline.

RESULT

Stale replay: receipt-only makes nine resumptions, eight unsupported, yet finishes placement; revalidation resumes once.

OBSERVATIONS

Revalidation resumes at 5.142/5.302 s for gap/replay; forced opening aborts at 5.442 s without resuming.

CONCLUSION

Final success can hide unsupported recovery. Delivery, a valid grasp and an appropriate path are separate conditions.

LIMITATIONS

Fixed privileged simulation, one clock and one run per cell; policy bundles change both gate and path. No regrasp, vision, ROS2 or hardware safety claim.

REPRODUCE

18 new and 41 existing tests pass. Audit covers 18 rollouts, 12 identical prefixes, two recovery windows and 108 raw files.

1. Engineering cases

   Embodied practice 5: communication is back—may transfer resume?

   18 MuJoCo rollouts reveal unsupported resumptions hidden by eventual success; validate fresh evidence and replan the remaining path.

   2026-09-3012 MINPublished
2. LAB / 008

   E4 / Observation freshness and sampling

   Separate delivery from fresh evidence and test a fixed count at different sampling periods.

   2026-09-29Completed
3. PROJECT / 003

   Hohoo Embodied Experiments

   Seven MuJoCo teaching experiments, from pickup bias to stage contracts, with raw trajectories, decisions and independent audits.

   2026-09-28Experiment