Reproduce “Task-Specified Active Metrological Inspection with Measurement-Steered VLA Manipulation and Deterministic Evidence Gating” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Zhiling Chen, Jingzhan Ge, Ruimin Chen, Matthew P. Castanier, David Gorsich, Farhad Imani

FRAME is an autonomous inspection framework for high-mix, low-volume manufacturing that converts natural-language requests and structured specifications into traceable metrological decisions. By combining learned visual grounding and manipulation with deterministic scope, correspondence, evidence-admission, and coverage checks, it reduced wrong correspondence releases to 2.4% on L1 parts and 3.6% on L2 parts, while abstaining when valid evidence was insufficient.

Key insights

  • The framework treats evidence validity and requirement coverage as prerequisites for conformance, separating physical task completion from trustworthy inspection outcomes.
  • Deterministic checks constrain learned components: the VLM selects specification identifiers and correspondence candidates, but does not supply tolerances or independently authorize measurements and releases.
  • Conservative abstention improves safety under ambiguity. With FRAME, released correspondence accuracy reached 97.4% for L1 and 95.6% for L2 parts, while harder cases produced higher abstention rather than more false releases.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.