gazebo-operator

Controlling a running Gazebo simulation via natural language involves inspecting the active world to resolve target entities, selecting the correct topic or service interface, generating schema-correct gz.msgs payloads, and executing the action with post-execution verification.

Setting a model's pose or velocity in Gazebo without raw commands requires a schema-aware system to match your intent to the target entity, select the relevant transport interface, generate the correct gz.msgs payload, and verify the resulting state change.

The safest way to pause, resume, or step a Gazebo simulation is by routing intent through bundled scripts that enforce safety rules, preventing direct raw gz command composition and verifying the transport acknowledgements and simulation state after execution.

Discovering available topics, services, and entities in an active Gazebo world requires runtime inspection and capability discovery, which queries the running simulation to map available interfaces and entities before resolving the target for any action.

You can extend Gazebo control actions for custom world or topic patterns by adding Python adapters under the adapters directory, allowing custom action routing when fixed actions like pausing, stepping, or setting pose are insufficient.

Composing raw gz commands for Gazebo simulation operations is unsafe because it bypasses schema-aware payload generation and interface resolution, risking incorrect message structures and unverified execution outcomes that bundled safety scripts are designed to prevent.