Repository navigation
docs(action) clarify AgiBotWorld URDF scope - #338
sylvesterkaczmarek wants to merge 1 commit into
Conversation
fa79d94 to
feeba78
Compare
|
Refreshed this docs PR onto current main today. It is now 0 commits behind and mergeable, and the diff remains limited to the intended documentation file. This has been quiet for over two weeks. Could a maintainer please review/merge it, or let me know if any wording change is needed? |
Refreshed onto current upstream. Signed-off-by: Sylvester Kaczmarek <16242628+sylvesterkaczmarek@users.noreply.github.com>
feeba78 to
10894fe
Compare
|
Could a maintainer review the AgiBotWorld URDF scope clarification? The PR and its validation workflow are still waiting for review and approval. |
|
Thanks for this, @sylvesterkaczmarek. The Cosmos cookbooks and evaluation suites have moved to NVIDIA/cosmos-framework under The AgiBotWorld URDF scope note (#305) isn't in Cosmos Framework yet. Could you re-submit it against The relative paths are unchanged, just under |
Carries over the two documentation clarifications from NVIDIA/cosmos#334 and NVIDIA/cosmos#338 after the cookbooks moved into Cosmos Framework. - Explain that supported action widths depend on registered embodiment/domain definitions and compatible model weights, including dataset-defined exceptions. Addresses NVIDIA/cosmos#220. - Explain that the published AgiBotWorld omnipicker URDF is an FK/kinematics reference; its mesh dependencies are absent, and the FK loader removes visual/collision elements. Addresses NVIDIA/cosmos#305. Both original PRs were closed solely because the files migrated to `.agents/skills/`; this new PR preserves the original explanations at their new location. Validation: Documentation-only change. Links to the domain registry and URDF were verified against upstream, and `git diff --check` passed. No runtime or model execution claimed. DCO signed off. Thanks @yaoxu-crypto for the clear migration guidance. Signed-off-by: Sylvester Kaczmarek <16242628+sylvesterkaczmarek@users.noreply.github.com> Co-authored-by: pengcuo <pzeren@nvidia.com> Co-authored-by: yaoxu-crypto <yaoxu@nvidia.com>
Summary
G1_omnipicker_calibrated.urdfas a kinematics referencerobot_assetsWhy
Issue #305 asks why the AgiBotWorld Beta data appears to use a CTEK gripper while the available URDF is an omnipicker description, and asks for the complete mesh-backed URDF.
The current public Cosmos Framework hard-codes
G1_omnipicker_calibrated.urdffor the AgiBotWorld FK path. The URDF containspackage://genie_robot_description/meshes/...references, but the corresponding mesh package is not included in the publicrobot_assetsdirectory. This does not block the current action FK path because the loader explicitly removes<visual>and<collision>elements before loading the model into MuJoCo.This documentation change makes the scope of the published robot description explicit and avoids implying that it is a complete renderable CTEK/omnipicker asset package.
Validation
Documentation-only change. Cross-checked against
agibot_spec.py,agibot_fk.py, the committedG1_omnipicker_calibrated.urdf, and the currentrobot_assetsdirectory in NVIDIA/cosmos-framework.Addresses #305