URML (open robot intent language): a validate-before-actuate layer above Magma — request for comment

[idoco2003]

Hi, and thanks for Magma. What drew me here is that Magma grounds language and vision into action, UI actions and robot manipulation, so the output is a concrete action rather than just text.

I maintain URML, an open (Apache-2.0) language for declaring robot intent that sits one layer above a policy like Magma. A robot declares a capability manifest (grippers, reach, payload, sensors) and a safety envelope, and URML validates each action against that declaration before it executes. The model proposes; a static checker refuses anything the robot cannot do or is not allowed to do, then lets the rest through. Validate before actuate.

For a foundation model that emits robot action, that gate is the natural complement: Magma decides what to do, URML checks the specific action is admissible on the specific robot before it reaches hardware. It does not constrain the model; it makes the action surface checkable.

Two questions:

1. Would a declared capability and safety envelope be a useful guardrail on top of Magma's action output, especially for manipulation across embodiments?
2. Would a small worked example mapping a Magma action to a URML manifest (validated, no execution) be worth having, in your examples or ours?

Disclosure: URML's prose and examples are AI-assisted under my direction, disclosed openly in our VIBE.md. Happy to go deeper.

Thanks,
Ido (URML, urml.dev)