Hi dex-retargeting community,
URML (urml.dev) is a small, Apache-2.0 language for describing robot intent: it validates a request against a capability manifest and a safety envelope, then dispatches. dex-retargeting is the human->robot hand retargeting layer behind AnyTeleop, used by many VR/vision teleop rigs, and URML is interesting to it as the validated bound around the retargeted output.
Nothing here asks the project to adopt, host, or maintain anything. This is a request for comment, and a design question.
A retargeted grasp still has to be something the declared hand can actually do -- DOF, joint limits, the object's graspable class -- and that's exactly URML's capability/envelope check. So a retargeting result becomes (or is paired with) a URML manipulation intent that is validated against the declared dexterous-hand manifest before it's sent to hardware. This connects to dexterous-hand manifest questions URML has been asking elsewhere (LEAP / Shadow): what a multi-DoF hand must declare so a retargeted grasp can be capability-checked.
Two real questions: (1) Is wrapping a retargeted hand pose in a validated manipulation intent (capability + envelope check before hardware) interesting for AnyTeleop-style rigs? (2) What should a URML capability manifest declare to describe a dexterous hand honestly so a retargeted grasp can be checked (DOF, joint limits, graspable classes)?
Full write-up: https://github.com/URML-MARS/URML/blob/main/docs/rfcs/0481-dex-retargeting-outreach.md
Thanks for dex-retargeting; the retargeting layer most VR teleop rigs depend on is the right place to think about a validated bound on the result.
Ido Yahalomi (URML, greenvh@gmail.com)
AI-assisted prose, maintainer-reviewed before posting (see VIBE.md). Human-only correspondence available on request.
Hi dex-retargeting community,
URML (urml.dev) is a small, Apache-2.0 language for describing robot intent: it validates a request against a capability manifest and a safety envelope, then dispatches. dex-retargeting is the human->robot hand retargeting layer behind AnyTeleop, used by many VR/vision teleop rigs, and URML is interesting to it as the validated bound around the retargeted output.
Nothing here asks the project to adopt, host, or maintain anything. This is a request for comment, and a design question.
A retargeted grasp still has to be something the declared hand can actually do -- DOF, joint limits, the object's graspable class -- and that's exactly URML's capability/envelope check. So a retargeting result becomes (or is paired with) a URML manipulation intent that is validated against the declared dexterous-hand manifest before it's sent to hardware. This connects to dexterous-hand manifest questions URML has been asking elsewhere (LEAP / Shadow): what a multi-DoF hand must declare so a retargeted grasp can be capability-checked.
Two real questions: (1) Is wrapping a retargeted hand pose in a validated manipulation intent (capability + envelope check before hardware) interesting for AnyTeleop-style rigs? (2) What should a URML capability manifest declare to describe a dexterous hand honestly so a retargeted grasp can be checked (DOF, joint limits, graspable classes)?
Full write-up: https://github.com/URML-MARS/URML/blob/main/docs/rfcs/0481-dex-retargeting-outreach.md
Thanks for dex-retargeting; the retargeting layer most VR teleop rigs depend on is the right place to think about a validated bound on the result.
Ido Yahalomi (URML, greenvh@gmail.com)
AI-assisted prose, maintainer-reviewed before posting (see VIBE.md). Human-only correspondence available on request.