ACT Policy - SO-100 Pick and Place

Model trained on 50 episodes of pick-and-place task with SO-100 robotic arm.

Training Details

  • Steps: 20,000
  • Dataset: tomduf70/so100_first_dataset
  • Architecture: ACT (Action Chunking Transformer)
  • Vision Backbone: ResNet18
  • Final Loss: ~0.089

Hardware

  • Robot: SO-100 (Follower/Leader setup)
  • Cameras: 2x (top view + side view, 640x480 @ 15fps)

Usage

from lerobot import LeRobotPolicy

policy = LeRobotPolicy.from_pretrained("tomduf70/act_so100_v1_checkpoint_20k")

Or with CLI:

lerobot-record --policy.path=tomduf70/act_so100_v1_checkpoint_20k ...
Downloads last month
8
Video Preview
loading