VLA serving engine in OmniBot

The vision-language-action serving code is a folder in the OmniBot repository, not a standalone product that has been released on its own.

vla_engine is a FastAPI wrapper around an OpenVLA-style model. A ROS 2 node can call it. Weights are not bundled, and a GPU with enough memory is required before inference does anything useful. There is no reliability number published for it.

The folder on the public repo is omnibot-ai-engines/vla_engine. Contributions belong on that repository once you have cloned it.