Add MicroAGI embodiment with data configs; fix viz_language image-key resolution#495
Draft
marcopepunkt wants to merge 2 commits into
Draft
Add MicroAGI embodiment with data configs; fix viz_language image-key resolution#495marcopepunkt wants to merge 2 commits into
marcopepunkt wants to merge 2 commits into
Conversation
added 2 commits
June 11, 2026 21:08
- Microagi Human-sibling embodiment with keymaps, transform list,
intrinsics fallback, MediaPipe / MANO keypoint connectivity
-microagi_bimanual / right_arm / left_arm embodiment IDs + hydra data/viz configs
- CONTRIBUTING_DATA.md embodiment table updates
Co-authored-by: Aristotelis-Sib <aristotelis98@gmail.com>
The evaluator/viz configs (keypoints.yaml, keypoints_wrist.yaml) specify image_key as a flat name like 'front_img_1' because they were written for the trainer's EvalVideo path, where HPT flattens dotted batch keys to their last segment before viz. viz_language operates on raw dataloader batches, where the same image is keyed 'observations.images.front_img_1', so those configs raised KeyError on every batch and produced zero videos. Resolve a missing image_key by matching batch keys on their last dotted segment — the inverse of HPT's flattening. Only the image key is resolved (wholesale flattening would collide, e.g. left/right.obs_keypoints), and only when the exact key is absent, so configs using full dotted keys (cotrain_lang.yaml) are unaffected.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Registers MicroAGI egocentric capture as a new Human embodiment and
fixes a key-naming mismatch that broke
viz_language.pywith the existingkeypoint viz configs.
New embodiment:
microagi_*(IDs 15–17)microagi_keypoints.yaml, mirrorsaria_keypoints.yaml)and viz entries for
evaluator/viz/keypoints*.yamlCONTRIBUTING_DATA.md: embodiment ID tableFix:
viz_language.pyresolved zero videos withkeypoints*.yamlThose viz configs use flat image keys (
front_img_1) written for the trainerpath, where HPT flattens dotted batch keys.
viz_language.pyreads rawdataloader batches (
observations.images.front_img_1), so every batch raisedKeyErrorand runs silently wrote 0 videos. Missing image keys are now matchedby their last dotted segment — the inverse of HPT's flattening, scoped to the
image key only so dotted-key configs (
cotrain_lang.yaml) are unchanged.Validation
unit-norm quaternions, 30 fps timestamps, JPEG q85, annotations)
inverse(obs_head_pose)+ intrinsics: skeletons land on the handspipeline — identical overlay behavior