E6 / Recovery gate and path ablation

01 / QUESTION

Separate revalidation from replanning: which differences concern evidence and which concern the resumed target?

METHOD

A 2×2 gate/path design plus a latched reference across six conditions: 30 rollouts, one per cell. Another 18 E5 regression runs are not extra E6 samples.

RESULT

Across four same-gate pairs, replanning reduces first-resume target jumps from 50.285–97.055 mm to 0.245–0.263 mm. All active combinations place in five non-drop conditions; no placement-rate improvement is observed.

OBSERVATIONS

With forced opening, receipt-only replanning/wallclock resume 72/6 times, all unsupported. Transfer-only monitoring and retiming both affect the difference.

CONCLUSION

Gating establishes recovery evidence; replanning improves target continuity in this matrix. Continuity does not establish safety or task success.

LIMITATIONS

Fixed scene, one clock, privileged simulation and one deterministic run per cell. Geometry and timing change together; no vision, regrasp, ROS2 or hardware validation.

REPRODUCE

30 local unit tests pass. Independent auditing checks 180 episode files, 12 source fingerprints, 36 identical prefixes and 102 resumptions. Original Linux runs and Windows revalidation remain distinct.

1. Engineering cases

   Embodied AI VI: what do recovery gating and replanning each solve?

   A 30-rollout ablation reduces first target jumps from 50–97 mm to about 0.25 mm, without improving placement outcomes in this fixed matrix. Separate recovery evidence, command continuity and task success.

   2026-09-3014 MINPublished
2. 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
3. LAB / 009

   E5 / Revalidating recovery

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

   2026-09-30Completed
4. 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