rosetta_interfaces

ROS 2 action and service definitions for Rosetta.

These form the control plane (start/stop/manage). The data plane (observations, actions, rewards) is plain ROS 2 topics defined by the contract YAML.

Interface	Kind	Server	Purpose
RecordEpisode	action	episode_recorder	Record a demonstration to a rosbag
RunPolicy	action	policy_runner	Run policy (or classifier) inference
ManageEpisode	action	hil_manager	Orchestrate a human-in-the-loop episode
StartHILEpisode	service	hil_manager	Start a HIL episode (returns once started)
StartRecording	service	episode_recorder	Lightweight start of recording
StartPolicy	service	policy_runner	Lightweight start of policy inference

The services exist because some clients (Foxglove, for one) can call services but not actions. They are thin proxies for the action operations they are named after, and they return immediately, per the ROS 2 design guidance that services "should never be used for longer running processes."

Every action result reports how it ended, not whether it "succeeded":

• The goal's terminal GoalStatus is the mechanics: SUCCEEDED when the work reached a defined end, CANCELED when a client took it away, ABORTED when the server stopped it (an error, or a lifecycle deactivate).
• termination_reason names the exact cause, with the legal values declared as constants in each .action.
• ManageEpisode additionally carries outcome (whether the robot did the task), which is independent of how the episode ended.

Field-by-field definitions with example calls: Actions and services reference. The .action and .srv files in this repo are the source of truth: they are commented, and short enough to read directly.

Build

colcon build --packages-select rosetta_interfaces
source install/setup.bash

Documentation

Full Rosetta documentation: https://iblnkn.github.io/rosetta/

License

Apache-2.0