01
Body
A compact two-axis head, 3D-printed mechanics and servo actuation.
Experiment 001 · AI HEAD / AI CUBE
AI_CUBE
Local embodied intelligence
AI_CUBE is a small physical AI experiment exploring what happens when language, perception and intent become real motion.
The system combines a local language model with deterministic motion control, audio, vision and a simple two-axis robotic head. The AI can decide what it wants to express, but the physical layer remains constrained, testable and safe. The project is deliberately small: a test bench for understanding how agentic software should meet real hardware.
02
Brain
Local AI interprets language and chooses intent without sending the interaction to the cloud.
03
Motion
Intent becomes deterministic capabilities such as look, nod, shake, glance and curious motion.
04
Purpose
A practical laboratory for embodied agents, safe autonomy and AI-assisted engineering.
System architecture
How it works
AI_CUBE deliberately separates meaning from motion. The local model may interpret a situation and choose an intent, but it never drives the servos directly. Every physical action passes through a small deterministic layer that keeps the behavior understandable, repeatable and constrained.
01
Listen / See
Audio and vision turn the physical world into observations the system can reason about.
02
Local AI
The model understands context, language and intent while remaining on the local machine.
03
Agent Action
The AI chooses a published semantic action such as nod, look, glance or curious.
04
Guard
Deterministic validation checks that the requested action and parameters are allowed.
05
Capability
A known core capability converts intent into a deterministic motion sequence.
06
Executor
The servo layer performs the motion and the physical result becomes observable again.
AI decides what. Deterministic software decides how.
Observation → Intent → Action → Guard → Capability → Motion