← News

ResearchSources: arXiv · 18 Nov 2026

Holding reference choice can flip rankings of visual recovery methods, study finds

An arXiv abstract introduces AuditEmbodied, a paired evaluation framework showing that how a robot holds its pose during a wait can reverse which recovery method performs best.

In plain words

Researchers studying how robots recover from mistakes found that the outcome depends on how the robot's position is referenced while it waits. When they changed that reference, the better-performing method swapped places with the worse one. They propose treating this reference choice as an explicit condition in any fair comparison.

What was announced

The authors describe AuditEmbodied as a paired execution and trajectory evaluation framework that varies holding references and waiting duration independently while keeping the recovery executors frozen. A 960-execution factorial across 20 initialization seeds compares CAD-feedback and landmark-based recovery from identical post-wait physical states under each holding and waiting condition. Protected completion is defined as successful recovery under the declared execution constraints.

The reported effect is large: at a 1 s wait with registered wrist references, protected completion counts shift from 6/40 versus 19/40 under registered holding to 34/40 versus 22/40 under position-only holding. Trajectory analysis traces the main separation to grasp formation, where small observation-time target residuals precede large tracking errors. On fresh seeds, gradual position reanchoring improved CAD recovery's long-wait completion by 42.5 percentage points and cut mean complete recovery time by 6.06 s relative to registered holding. Further tests with pretrained policies, handover, and a mass-matched geometry probe sensitivity across executors and physical regimes.

What it does not show

  • The evidence is an abstract only; full methodology, raw results, and peer review status are not available in the supplied passage.
  • All measurements and the reanchoring improvement are author-reported on their own benchmark and are not independently confirmed.

Sources

Source-support review completed 19 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.

Coverage: arXiv: abstract.

Updates and corrections

No corrections so far.

First published 19 Nov 2026

New to the subject? AI from the Ground Up starts from the beginning.