Running a Real Policy

The earlier zero-action example keeps the robot still, so its success rate stays at zero. In this section, you replace that policy with a pre-trained model. Each guide shows how to start the model server and run a saved YAML configuration.

Start with OpenPI for an interactive first rollout, then try GR00T. The VLM agent guide uses a hosted vision-language model as the policy, so it needs an inference API key instead of a model server.

• OpenPI
• GR00T
• VLM Agent