Components

RoIS 2.0 defines 17 basic HRI Components. This page shows how each maps to a physical robot and to a virtual avatar, and what OpenRoIS provides for it today.

Robots and Avatars Share Most Components

Perception and speech components run the same models whether their input comes from a robot camera or a webcam. Of the 17 basic components, 13 keep the same interface and the same underlying models across paradigms, differing at most in their input source, output device, or coordinate system. Only 4 (Reaction, Navigation, Follow, and Move) need implementations specific to the paradigm, which is precisely the boundary that per-backend components keep separate. This is what makes an adapter for a new platform inexpensive to write.

Component	Physical robot	Virtual avatar	Across paradigms
Person Detection	Detector on the robot camera	Detector on a webcam or virtual sensor	Identical
Person Localization	Depth camera and tracker	World position or webcam depth	Coordinate system differs
Person Identification	Face or re-identification model	Same model	Identical
Face Detection	Face detector	Same detector	Identical
Face Localization	Face mesh	Same face mesh	Identical
Sound Detection	Voice activity detection	Same detection	Identical
Sound Localization	Microphone array	Microphone array or virtual source	Source differs
Speech Recognition	Speech-to-text model	Same model	Identical
Gesture Recognition	Pose model	Same model	Identical
Speech Synthesis	Text to speech on a speaker	Text to speech with lip sync	Output differs
Audio Streaming	Microphone to WebRTC	Synthesized audio to WebRTC	Source differs
Video Streaming	Camera to WebRTC	Rendered frames to WebRTC	Source differs
System Information	Battery, pose, and health	Frame rate and avatar state	State differs
Reaction	LEDs or gestures	Animations and expressions	Paradigm-specific
Navigation	Navigation stack, for example Nav2	Navigation mesh	Paradigm-specific
Follow	Navigation and tracking	Virtual follow behavior	Paradigm-specific
Move	Velocity commands to motors	Avatar transform	Paradigm-specific

Status in OpenRoIS

Typed models are the per-component Python, TypeScript, and C# types generated by the type pipeline. Reference implementations are published components that run against real hardware.

Component	Typed models	Reference implementation
System Information	Available	Preferred Robotics Kachaka (gRPC, ROS 2)
Navigation	Available	Preferred Robotics Kachaka (gRPC, ROS 2)
Person Detection	Available	In Progress (Intel RealSense, PR #3)
Reaction	Available	Planned
Person Localization	In Progress (PR #3)	In Progress (PR #3)
Person Identification	In Progress (PR #3)	In Progress (PR #3)
Speech Synthesis, Speech Recognition	Planned	Planned
Face Detection, Face Localization	Planned	Planned
Sound Detection, Sound Localization	Planned	Planned
Gesture Recognition	Planned	Planned
Follow, Move	Planned	Planned
Audio Streaming, Video Streaming	Planned	Planned (control plane demonstrated by the mock adapter's VideoStreaming)

A reference platform based on the open-source Pollen Robotics Reachy Mini is in progress (Phase 8 of the roadmap), so that every basic component can be tried on affordable, openly documented hardware. Its adapter covers System Information, Navigation (head and body poses), Reaction, Speech Synthesis, Speech Recognition, Person Detection, Face Detection, and Video Streaming, and runs in simulation today.

User-Defined Components

RoIS allows components beyond the basic 17, reusing the common messages and the profile mechanism. OpenRoIS supports them with the same decorators: declare the component name and its RoSO function classification, for example @component("NavigationInformation") for a component that publishes the destinations a robot knows. Service applications discover user-defined components through the profile exactly like basic ones.

Community Components

Adapters for additional platforms are developed as community components, separate from the core repository. A catalog and registry for them is part of the long-term roadmap.