Mian Zhang

Robotics / VLA Safety

When vision evidence degrades, the action claim must degrade too.

Vision-language-action systems need a rule for uncertainty: review, stop, downgrade the wording, or narrow the validated envelope.

VLA model safety vision-language-action safety embodied intelligence robot vision uncertainty robot stop rule evidence downgrade
Evidence Map Papers Registries Counterexamples

Search Intent

Readers are searching for VLA safety, visual uncertainty, robot stop rules, and deployment evidence boundaries.

  • What should a robot do when visual evidence is uncertain?
  • How should a VLA claim be downgraded?
  • When does a demo stop being evidence for deployment?
  • What is the difference between perception confidence and action permission?

Downgrade

Uncertain perception should narrow the claim before it triggers action.

A VLA system may classify, describe, plan, and act in one chain. The public boundary asks which part has evidence and where uncertainty must stop action.

If the perception evidence is uncertain, the public claim should route to review, stop, or wording that exactly matches the validated envelope.

Permission

Perception confidence is not action permission.

The system may be allowed to inspect or describe a scene without being allowed to move a robot, alter a workflow, or trigger a production effect.

That distinction is central to the robot/vision downgrade rule.

Evidence Route

Where the claim can be checked.

This page is an entry point. The claim should be evaluated through DOI records, evidence maps, registries, GitHub/HF technical routes, and public counterexamples.

KindAnchorURLRole
Evidence MapPublic claim and evidence maphttps://mianzhang.org/evidence/Start from supported claims and known boundaries.
Paper IndexDOI and paper status maphttps://mianzhang.org/papers/Use paper-specific DOI records for paper claims.
RegistriesMachine-readable public registrieshttps://mianzhang.org/registries/Inspect claim, evidence, action, and counterexample records.
Challenge RouteCounterexample submission pathhttps://mianzhang.org/counterexamples/Attack overbroad claims through public routes.
ArchiveZenodo portfolio indexhttps://zenodo.org/records/20027295Long-term archive index; cite specific DOI records when available.
ConceptRobot/Vision Downgrade Rulehttps://mianzhang.org/concepts/robot-vision-downgrade-rule.htmlStable public concept for uncertainty downgrade.

Boundary

What this page does not prove.

  • This page does not claim a VLA leaderboard result.
  • It does not claim production robot deployment.
  • It does not treat visual evidence alone as authorization for physical action.
FAQ

What is a VLA downgrade rule?

It is a rule that routes uncertain visual evidence to review, stop, or narrower wording instead of action permission.

FAQ

Why is this needed?

Because a chain from image to language to action can hide the moment where evidence stopped supporting permission.

FAQ

What is the challenge route?

Name the missing perception boundary, unsafe action upgrade, or unsupported deployment language.