gamzadole/omx_f_box_stacking_act_v1
ACT (Action Chunking Transformer) policy trained with LeRobot.
Task
Pick-and-place with an OpenManipulator-X (omx_f) arm.
Note: the source dataset labels the task as
"Stack box on the green zone", but the actual demonstrated behavior is a pick-and-place motion. The label is inherited from data collection and does not affect the policy (single-task dataset).
Training
- Dataset:
gamzadole/omx_f_Box_Stacking_v2(repaired locally — 8 episodes had missing files andmeta/tasks.jsonlwas absent; usable data: 43 episodes / 15,165 frames) - Steps: 80,000, batch size 16
- Final training loss ~0.035
Usage
from lerobot.policies.act.modeling_act import ACTPolicy
policy = ACTPolicy.from_pretrained("gamzadole/omx_f_box_stacking_act_v1")