act_panda_pick_v1

ACT fine-tuned on Rakeshdharan56/pick_robosuite (revision v1): Franka Panda, "Hand me the {bottle|vial|syringe}", LIBERO-style I/O (2 cams 256², 8-D state, 7-D delta-EEF action at 20 Hz).

Non-VLA baseline trained from scratch (not language-conditioned), 50,000 steps, batch 64, chunk 50 / n_action_steps 50, RTX 6000 Ada. No camera rename: observation.images.agentview, observation.images.eye_in_hand.

Load with LeRobot 0.5.2 (commit b06ad4088): get_policy_class(cfg.type).from_pretrained("Rakeshdharan56/act_panda_pick_v1") and make_pre_post_processors(policy.config, pretrained_path="Rakeshdharan56/act_panda_pick_v1"). Closed-loop robosuite eval: pending.

Downloads last month
-
Safetensors
Model size
51.6M params
Tensor type
F32
·
Video Preview
loading

Dataset used to train Rakeshdharan56/act_panda_pick_v1