# Preprint presents method for planning with uncertain AI perception

_Published Saturday, November 14, 2026 at 6:06 AM EST · AI, Science · Latest · Tier 2 — Notable_

Researchers present a method in a preprint that improves downstream task planning by retaining several plausible states from an AI perception model instead of treating its single best guess as certain.

The method uses the model's existing prediction probabilities to construct the smallest set of states meeting a user-selected probability threshold. Its search algorithm, PRaVaSS, combines per-feature predictions with task constraints while keeping a compact representation of uncertainty. The researchers report validation of scalability and improved decisions in a task-planning pipeline using a vision-language model. The algorithm's polynomial complexity guarantee applies to the unconstrained case.

## Sources

- [cs.AI updates on arXiv.org](https://arxiv.org/abs/2611.14781)

---
Canonical: https://techandbusiness.org/newswire/ac_fap2lg_NPVmmx7MiEqs
Published: 2026-11-14T11:06:21.868Z
Story chronology: 2026-11-14T05:00:00.000Z
Retrieved: 2026-11-15T00:01:48.219Z
Publisher: Tech & Business (techandbusiness.org)