Hello Unitree Robotics / UnifoLM_WBT maintainers,
We reviewed the current official Hugging Face README and found useful partial documentation, but several action-target semantics remain unresolved before any offline lower-layer binding. Could you please answer the following targeted questions or point us to an official source?
- For UnifoLM_WBT action.ee_action, the current official README only partially covers source_unit_and_scale_evidence: official README states action.ee_action is target end-effector state from FK rooted at the root link. Please provide an authoritative answer for the remaining point(s): official units for each of the 12 action.ee_action values; official scale convention for position and orientation values.
- For UnifoLM_WBT action.ee_action, the current official README only partially covers coordinate_frame_or_joint_order_evidence: official README gives root-link FK frame context and both-end-effector pose context. Please provide an authoritative answer for the remaining point(s): official 12D left/right order for action.ee_action; official pose component order for each end-effector; official orientation representation for action.ee_action.
- For UnifoLM_WBT action.hand_cmd, the current official README only partially covers source_unit_and_scale_evidence: official README states Inspire/BrainCo/Dex1 hand_cmd ranges and open-to-close direction. Please provide an authoritative answer for the remaining point(s): Dex5 applicability or range evidence if Dex5 is present.
- For UnifoLM_WBT action.hand_cmd, the current official README only partially covers coordinate_frame_or_joint_order_evidence: official README states per-hand command order for Inspire, BrainCo, and Dex1. Please provide an authoritative answer for the remaining point(s): left/right concatenation order for action.hand_cmd; Dex5 per-hand order evidence if Dex5 is present.
We will keep Phase 11 reference generation, action-target binding, IL/RL training, and WBC execution disabled until authoritative evidence is available and validated.
Context for this request:
- Operator identity for this offline audit packet: bumblebee
- Draft generated at: 2026-06-05T12:32:08.968447+00:00
- Target repository observed: unitreerobotics/unifolm-world-model-action
- Target repository URL: https://github.com/unitreerobotics/unifolm-world-model-action
- Purpose: clarify offline structured-intent action-target semantics before any lower-layer binding.
- This request does not claim robot execution, action prediction, IL/RL training, WBC execution, or closed-loop recovery success.
Hello Unitree Robotics / UnifoLM_WBT maintainers,
We reviewed the current official Hugging Face README and found useful partial documentation, but several action-target semantics remain unresolved before any offline lower-layer binding. Could you please answer the following targeted questions or point us to an official source?
We will keep Phase 11 reference generation, action-target binding, IL/RL training, and WBC execution disabled until authoritative evidence is available and validated.
Context for this request: