Proposal: reproduce one DuoBench stage-contract task in OmniSim #4

Hi Robot Control Stack team,

I maintain OmniSim, an Apache-2.0 robotics simulator with agent-facing HTTP/MCP control, URDF import, measured manipulation examples, and CPU/GPU Newton physics.

DuoBench's stage-level task contract is especially useful because it gives more diagnostic information than a binary success flag. We would like to reproduce one small DuoBench task in OmniSim while preserving the public Gymnasium-facing semantics:

• stage, max_stage, success, and subinstruction progression;
• joint-space action timing and binary gripper behavior;
• reproducible resets plus measured physical success/failure checks;
• exact scripts and asset-license/provenance notes.

We would do the OmniSim-side implementation and keep it separate until there is a result worth reviewing. We are not asking the DuoBench team to support another simulator.

Which existing task has the smallest asset and control surface for a credible first port?

Best,
Ahmed Fetouh — OmniLink

Disclosure: this note was drafted with the same AI-agent workflow used to build OmniSim and reviewed/sent by me.