Fine-tuning SmolVLA on an SO-101: what's your recipe?
The SO-101 pick-and-place set is a good template: about fifty episodes with a side and a top-down camera. Most people seem to start from the SmolVLA base checkpoint.
How many episodes did you record before the policy got reliable? Where did you put the cameras, and how long did training take on your GPU?
Referenced repos
Indexed from Hugging Face and linked upstream, not hosted here.
A compact 450M-parameter VLA from Hugging Face, pretrained on community-collected LeRobot datasets and sized for fine-tuning and inference on consumer GPUs.
Fifty real pick-and-place episodes from a low-cost SO-100/SO-101-class arm, recorded at 30 fps with a side camera and a top-down camera, and used as the SmolVLA fine-tuning example.
A browser front end for LeRobot that talks to a locally run FastAPI backend to calibrate, teleoperate, record, train, and replay policies on an arm without using the CLI.