Component Library

RoIS defines 17 basic HRI components. Every component (except System Information) shares the RoIS_Common interface: start, stop, suspend, resume, and component_status. About 70% of components are identical across paradigms. The perception and speech components run the same ML models whether the input is a robot camera or a webcam. Only actuation, world model, and stream source differ.

Component	Robot backend	Avatar backend	Shared?
Person Detection	YOLO on camera	YOLO on webcam	yes
Person Localization	depth + tracker	world position	diff coord system
Person Identification	InsightFace	InsightFace	yes
Face Detection	MediaPipe	MediaPipe	yes
Face Localization	MediaPipe face mesh	MediaPipe face mesh	yes
Sound Detection	mic VAD	mic VAD	yes
Sound Localization	mic-array DOA	mic-array DOA / virtual	diff
Speech Recognition	Whisper	Whisper	yes
Gesture Recognition	MediaPipe Holistic	MediaPipe Holistic	yes
Speech Synthesis	TTS to speaker	TTS to lip-sync	diff output
Reaction	LED / gesture	animation / expression	paradigm-specific
Navigation	Nav2 (physical)	NavMesh (virtual)	paradigm-specific
Follow	Nav2 + tracker	virtual follow	paradigm-specific
Move	cmd_vel to motors	transform to avatar	paradigm-specific
Audio Streaming	mic to WebRTC	TTS output to WebRTC	diff source
Video Streaming	camera to WebRTC	rendered frames to WebRTC	diff source
System Information	battery, CPU, joints	FPS, memory, avatar state	diff state

The component's logic is the same across adapters. Only the binding differs. The spec also supports user-defined components beyond the basic 17, reusing RoIS_Common and the profile mechanism. An HRI Component Profile can include another profile via sub_component, so an extended component can reuse a base component's messages and add new ones.