Skip to the content.

layout: post title: “From Tokens to Torques: Fine-Tuning a VLA on an OMX Manipulator” —

This is part one of my series on bringing Vision-Language-Action models out of simulation and onto a real 4-DoF OpenMANIPULATOR-X using Hugging Face and LeRobot.

### Loading the Policy

Here is how we pull the pretrained model:

```python from lerobot.common.policies.factory import make_policy

policy = make_policy( name=”my_vla_policy”, pretrained_name_or_path=”huggingface_username/omx_policy” )

Interactive 2-DoF Arm Demo

Adjust the sliders below to see how joint angles affect the end-effector position. This is a simple 2D representation of what our OMX manipulator does in 3D space!