Aurora Vision reads store shelves from a single wide-angle photo

What changed

• The vision model now handles wide-angle and angled shelf photos, correcting for perspective distortion instead of requiring a perpendicular shot.
• It flags likely out-of-stock gaps by comparing shelf layout against a planogram, and highlights products in the wrong section.
• Results now include a confidence score per flagged item so low-confidence guesses can be queued for a human check instead of auto-actioned.

Who it helps

The practical problem it solves

Manual shelf audits are slow and shots taken at an angle used to be unusable for automated checks. Accepting an ordinary angled photo turns a five-minute manual walk-and-note task into a ten-second photo with a ready-made gap list.

Ways to try it

• Snap one photo per aisle at closing time and get a same-night restock list.
• Spot-check planogram compliance across multiple locations without a specialized camera rig.
• Route only the low-confidence flags to a person, instead of re-checking every item.

More from this briefing

• Agent Sandbox can now replay a failed run step by step

  Coding agents

• Continuum adds human-approval checkpoints inside automated workflows

  Enterprise workflows