Partial fine-tune of GR00T N1.7 on the MolmoAct2-BimanualYAM-Dataset (3 cameras: top → left → right). Projector + diffusion trainable; LLM/vision frozen.

Checkpoint details

Setting	Value
Training step	30000
Cameras	top, left, right
Action dim	14 (absolute joints + grippers)
Chunk size	16
Embodiment	new_embodiment
Base model	nvidia/GR00T-N1.7-3B
Gripper range	[0, 1] (continuous)

Usage (LeRobot)

python

from lerobot.policies.groot.modeling_groot import GrootPolicy
from lerobot.policies.factory import make_pre_post_processors

ckpt = "Jiafei1224/molmoact2-yam-gr00t1.7"
policy = GrootPolicy.from_pretrained(ckpt)
preprocessor, postprocessor = make_pre_post_processors(policy.config, pretrained_path=ckpt)

Open-loop smoke (2 held-out episodes, stride 30, reset-each-step)

MSE: 0.00160
MAE: 0.0244