Download stack_blocks/meta/alignment.json from SWIRL-Lab/bigym-g1-native60: direct link, hf CLI and curl.
- Browser
- Download file 666 Bytes
-
https://huggingface.co/datasets/SWIRL-Lab/bigym-g1-native60/resolve/main/stack_blocks/meta/alignment.json
- Command line
-
hf download hf://datasets/SWIRL-Lab/bigym-g1-native60/stack_blocks/meta/alignment.json
-
curl -L -o alignment.json https://huggingface.co/datasets/SWIRL-Lab/bigym-g1-native60/resolve/main/stack_blocks/meta/alignment.json
666 Bytes
| {"action_alignment_version": 2, "transition_keys": ["action", "raw_outer_action", "expanded_action", "lowerbody_action", "leg_joint_targets", "torso_target", "lowerbody_command", "height_command", "reward", "event_progress", "discount", "demo", "is_expert"], "doc": "Frame k pairs the frame's observation with the transition executed FROM it (source npz index k+1) for every transition_keys feature; the final frame repeats the last real transition and source index 0 lives in episode_init_states.json first_transition. Datasets without this file are v1: features stored at source index k verbatim (one-frame hindsight misalignment for standard LeRobot consumers)."} |