Cosmo operates as a local-first, modular, event-driven robotic runtime. Its conceptual model is simple:
Perception -> Cognition -> Action
Perception gathers input from voice, future vision, sensors, system events, and local context. Cognition interprets that input against memory, personality, current state, and task intent. Action turns the response into speech, local automation, UI updates, or future movement through a physical robotics bridge.
Voice lifecycle
The current voice flow follows this sequence:
- Wake word
- Audio capture
- Speech-to-text
- Cognition
- Memory/context
- Text-to-speech
- Physical/system action
- Return to idle
Runtime states
Cosmo’s state machine keeps the system predictable:
- IDLE
- LISTENING
- PROCESSING
- SPEAKING
- COOLDOWN
Why state control matters
The purpose of explicit state control is to prevent duplicated wake word activations, overlapping speech, STT capturing TTS output, uncontrolled concurrent processing, and response generation before speech completion.
Cosmo should feel responsive without becoming chaotic. The state machine gives each subsystem a shared understanding of what the robot is doing before another action begins.