Skip to content

URML (open robot intent language): consuming a zoo model's estimate on a robot (request for comment) #312

Description

@idoco2003

Hi OpenCV Zoo maintainers,

URML (urml.dev) is a small, Apache-2.0 language for robot intent: an intent becomes a typed primitive, validated against the robot's declared capabilities and a safety envelope, then dispatched. opencv_zoo collects edge-deployable models for OpenCV DNN, and for the robotics subset of users, a zoo model's output is exactly the kind of perception fact a URML intent conditions on. This is a narrowly-scoped consume-the-estimate note about that robotics slice, not a claim on the whole project.

Nothing here asks the project to adopt, host, or maintain anything. This is a request for comment.

The boundary: for a robot running an opencv_zoo model on-device, the model's output (a detection, a pose, a segmentation) is a fact a typed URML intent conditions on, validated against the robot's capabilities and a safety envelope before dispatch. The zoo stays the model source and OpenCV DNN stays the runtime; URML stays out of perception. What a given zoo model outputs (its classes, keypoints) maps toward the perception side of a URML manifest, so an intent that needs an output the chosen model does not provide can be caught early.

Two real questions: (1) for robots using opencv_zoo models on-device, is "the zoo model produces the estimate, URML consumes it as a fact an intent conditions on" a sensible boundary? (2) Does a zoo model's output schema map cleanly toward the perception side of a URML manifest?

Full write-up: https://github.com/URML-MARS/URML/blob/main/docs/rfcs/0614-opencv-zoo-outreach.md

Thanks for the zoo; edge-deployable models are exactly what a lot of robots run for perception, and a clean line to the intent that uses them seemed worth raising for that subset.

Ido Yahalomi (URML, greenvh@gmail.com)

AI-assisted prose, maintainer-reviewed before posting (see https://github.com/URML-MARS/URML/blob/main/VIBE.md). Human-only correspondence available on request.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions