Hi safe-control-gym maintainers,
URML (urml.dev) is a small, Apache-2.0 language for describing robot intent: it validates a request statically against a capability manifest and a safety envelope, then dispatches. The envelope declares limits at the intent level (bounds, geofences, occupancy). safe-control-gym is where limits like those become control-theoretic constraints and a controller is actually evaluated for honoring them, which is the level below where URML sits.
Nothing here asks safe-control-gym to change or maintain anything. This is a request for comment on whether the two levels line up.
Two real questions. First, does a URML safety-envelope constraint set map cleanly onto safe-control-gym's constraint specification (state and input constraints, CBF formulations), or is the intent-level declaration too coarse to be useful at the control level? Second, could a benchmarked controller's constraint-satisfaction record sensibly inform a URML capability or envelope (this controller honors these limits, so declare them safe to rely on)?
Full write-up, with the constraint mapping: https://github.com/URML-MARS/URML/blob/main/docs/rfcs/0368-safe-control-gym-outreach.md
Thanks for safe-control-gym; a shared benchmark for safe control and safe RL is exactly what the field needed.
Ido Yahalomi (URML, greenvh@gmail.com)
AI-assisted prose, maintainer-reviewed before posting (see VIBE.md). Human-only correspondence available on request.
Hi safe-control-gym maintainers,
URML (urml.dev) is a small, Apache-2.0 language for describing robot intent: it validates a request statically against a capability manifest and a safety envelope, then dispatches. The envelope declares limits at the intent level (bounds, geofences, occupancy). safe-control-gym is where limits like those become control-theoretic constraints and a controller is actually evaluated for honoring them, which is the level below where URML sits.
Nothing here asks safe-control-gym to change or maintain anything. This is a request for comment on whether the two levels line up.
Two real questions. First, does a URML safety-envelope constraint set map cleanly onto safe-control-gym's constraint specification (state and input constraints, CBF formulations), or is the intent-level declaration too coarse to be useful at the control level? Second, could a benchmarked controller's constraint-satisfaction record sensibly inform a URML capability or envelope (this controller honors these limits, so declare them safe to rely on)?
Full write-up, with the constraint mapping: https://github.com/URML-MARS/URML/blob/main/docs/rfcs/0368-safe-control-gym-outreach.md
Thanks for safe-control-gym; a shared benchmark for safe control and safe RL is exactly what the field needed.
Ido Yahalomi (URML, greenvh@gmail.com)
AI-assisted prose, maintainer-reviewed before posting (see VIBE.md). Human-only correspondence available on request.