Hi Kenning maintainers,
URML (urml.dev) is a small, Apache-2.0 language for robot intent: an action becomes a typed primitive, validated against the robot's declared capabilities and a safety envelope, then dispatched. Kenning deploys and optimizes edge-AI models with ROS 2 CV node support; URML is the layer above the model's output that turns a perception or policy result into a validated action. This is a request for comment.
Nothing here asks the project to adopt, host, or maintain anything.
Two seams. First: Kenning gets the model running efficiently on the device, and URML consumes its output (a detection, a policy decision) as a fact a typed intent conditions on, validated before dispatch; Kenning keeps the deployment and optimization, URML stays out of inference. Second, and more interesting: Kenning produces a deployed, optimized model artifact, and URML has a direction (LearnedPolicy) where such an artifact could carry the operating envelope it is valid within, so an intent that relies on it is checked against that envelope. For an edge deployment where the model is quantized or otherwise optimized, knowing the envelope it still holds for is genuinely useful.
Two real questions: (1) is a typed, validated action layer downstream of a Kenning-deployed model useful? (2) Could a deployed/optimized model artifact carry an operating envelope a URML intent is checked against?
Full write-up: https://github.com/URML-MARS/URML/blob/main/docs/rfcs/0612-kenning-outreach.md
Thanks for Kenning; the deploy-and-optimize-for-edge step is exactly where the question of "what envelope does this optimized model still hold for" becomes concrete.
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.
Hi Kenning maintainers,
URML (urml.dev) is a small, Apache-2.0 language for robot intent: an action becomes a typed primitive, validated against the robot's declared capabilities and a safety envelope, then dispatched. Kenning deploys and optimizes edge-AI models with ROS 2 CV node support; URML is the layer above the model's output that turns a perception or policy result into a validated action. This is a request for comment.
Nothing here asks the project to adopt, host, or maintain anything.
Two seams. First: Kenning gets the model running efficiently on the device, and URML consumes its output (a detection, a policy decision) as a fact a typed intent conditions on, validated before dispatch; Kenning keeps the deployment and optimization, URML stays out of inference. Second, and more interesting: Kenning produces a deployed, optimized model artifact, and URML has a direction (LearnedPolicy) where such an artifact could carry the operating envelope it is valid within, so an intent that relies on it is checked against that envelope. For an edge deployment where the model is quantized or otherwise optimized, knowing the envelope it still holds for is genuinely useful.
Two real questions: (1) is a typed, validated action layer downstream of a Kenning-deployed model useful? (2) Could a deployed/optimized model artifact carry an operating envelope a URML intent is checked against?
Full write-up: https://github.com/URML-MARS/URML/blob/main/docs/rfcs/0612-kenning-outreach.md
Thanks for Kenning; the deploy-and-optimize-for-edge step is exactly where the question of "what envelope does this optimized model still hold for" becomes concrete.
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.