fine-tuning-openvla-oft
orchestra-research804 installs
Train robot policies for simulation or real arms.
What it does
Guides users through training, evaluating, and serving robot action policies with OpenVLA-OFT and OpenVLA-OFT+.
When to use it
Use for visually and language-conditioned robot action generation, particularly official LIBERO or ALOHA workflows with continuous action heads.
How to use it
Give it a robot-policy training or evaluation request; it sets up the official codebase and runs the corresponding workflow.
Uses
OO
AL
Latest review
What you need · 3
Requires conda to create and activate the training environment.
Requires GPU compute: evaluation lists one A100/A40 with about 16–18 GB VRAM; paper-default training lists eight A100s.
Access · 6
This skill
VLA server
Execute
Starts a robot inference server
Write
Requests actions from the server
ALOHA
Write
Controls your ALOHA robot