Vision Language Action Deployment

I deployed the OpenVLA model on top of MuJoCo and Robosuite for manipulation tasks. A custom position control API decodes the model output into end effector position, orientation, and grip, so the robot can pick and place objects directly from language and vision without a separate low level controller.