dp_isaaclab_ur7e_3task_fixed

Diffusion Policy (LeRobot diffusion) checkpoints trained on the Isaac Lab UR7e 3-task datasets (joint-space fixed data). Each subdirectory is a LeRobot pretrained_model folder at training step 60000 (no optimizer / training state included).

Subdirectory	Training dataset	Steps	Batch
grasp_pan/	DexSteer/isaaclab_ur7e_3task_fixed_grasp_pan	60000	16

Inputs: observation.images.third_person (3x480x640), observation.images.eye_in_hand (3x480x640), observation.state (24). Output: action (19). n_obs_steps=2, horizon=64, n_action_steps=32, ResNet18 backbone with separate encoder per camera.

Usage

from lerobot.policies.diffusion.modeling_diffusion import DiffusionPolicy

policy = DiffusionPolicy.from_pretrained("DexSteer/dp_isaaclab_ur7e_3task_fixed", subfolder="grasp_pan")