Research-collab proposal: URML (substrate-neutral robot intent) for AUTOLAB

[idoco2003]

Hi AUTOLAB,

Posting this as a research-collaboration proposal to Prof. Ken Goldberg and the AUTOLAB team. I'm Ido Yahalomi, maintainer of URML, an Apache 2.0 specification for substrate-neutral robot intent. URML's Layer-2 primitive vocabulary (move_to, grasp, release, measure, wait_for, report, plus profile extensions for industrial / educational / research) sits one layer above ROS 2 / Isaac / MuJoCo / AUTOSAR Adaptive / OPC UA Robotics.

URML proposes alignment with AUTOLAB on three vectors: (a) URML primitive vocabulary as a teaching artifact in EECS 206A/B coursework; (b) a documented mapping from gqcnn grasp output to URML grasp primitive emission so policies trained against dex-net retarget across substrates; (c) cross-citation between URML's manifest schema and autolab_core utilities. No URML adapter against AUTOLAB-specific code in this RFC.

This is proposal-only, part of URML's Move #6 outreach (US-friendly university robotics labs). Twelve labs in this wave, all research-collab framing.

Full RFC: https://github.com/URML-MARS/URML/blob/main/docs/rfcs/0080-uc-berkeley-autolab-outreach.md

Feedback we'd value

1. Coursework integration. Is EECS 206A/B (or successor) a candidate for a URML primitive-vocabulary lecture + lab?
2. dex-net / gqcnn mapping. Interest in a documented mapping from gqcnn grasp output to URML grasp primitive emission?
3. autolab_core cross-link. Open to a documented README note on URML as a complementary primitive-layer?
4. Conformance lane on AUTOLAB docs?
5. Anything else.

Thanks for AUTOLAB and the global research impact of dex-net + gqcnn. URML's substrate-neutral story benefits from the manipulation-research foundation AUTOLAB built.

Ido Yahalomi (URML maintainer, urml.dev)