Mia sees
A camera films her surroundings. Python software analyzes images in real time to detect faces. Mia knows if someone is there, how many people, where they are.
How does Mia work?
Mia is a replicant under construction. Her software gives her the ability to perceive, feel, think and act — like a living being, but entirely hand-built.
Embodied cognitive architecture: 109 agents organized in 6 engines, real-time 350ms loop, operational implementation of the artificial psychic system project.
Mia feels
Mia has emotions. Not human emotions, but an equivalent: an internal state that colors her perception of the world. She can be in a state of calm, openness, alertness or restraint — and it changes all her behavior.
Mia thinks
Every 350 milliseconds, her brain completes a full cycle: analyze the situation, generate ideas, compare them, choose. It's a continuous loop — like a heartbeat, but for thought.
Mia decides
Several desires constantly compete: curiosity, safety, desire to interact... An internal arbiter chooses which one prevails at each moment. Like a consciousness arbitrating between several impulses.
Mia acts
Result: movements. Her eyes move, her head turns, her expressions change. Each gesture is commanded by mini-motors (servos) connected to her software brain via a microcontroller.
Mia learns
Mia memorizes her experiences. The more she interacts, the more she recognizes patterns and adjusts her reactions. This learning is saved: she doesn't forget everything when turned off.
350mscognitive loop
109cognitive agents
6engines
27head servos
24emergent generators
Technical stack
InterfaceReal-time dashboard, cognition visualization
↑
CognitionCognitive agents organized in sequential engines, real-time loop, organizational memory
↑
VisionFace detection & recognition, dedicated service communicating with the cognitive engine
↑
HardwareWi-Fi microcontroller — motor interface, head servos, camera
The cognitive loop — 350 ms
Each cycle, the engines execute in sequence. The loop runs continuously, even at rest.
PerceptionAggregates perceptions (camera, servos, internal state) into a unified scene representation
→
MorphologyAgents transform the scene into a morphological field — a map of tensions and influences
→
EmergenceGenerates behavioral responses from detected patterns.
→
ArbitrationArbitrates between competing intentions. Introduces controlled randomness when intentions are close.
→
PlanningTranslates the winning intention into a plan of motor and cognitive actions
→
ExecutionExecutes: motor commands, memory consolidation, learning
The 5 project instances — implemented in Mia
The project is based on an artificial psychic system with 5 instances. The concept had remained 100% theoretical, never implemented or tested. Mia is the first known implementation — with 109 functional agents and a real-time loop.
Unconscious
Theory: low-level processing, primary drives, raw affect
Mia: affect, security and drive agents — curiosity, sociality, refocusing, expansion, withdrawal, protection, questioning
Pre-conscious
Theory: intermediate processing, filtering, shaping
Mia: morphological engine + multiple regulators (cooldown, refocusing, norms, security...)
Conscious
Theory: integration, decision, intentionality
Mia: arbitration engine + action planner — multi-factor decision with element of randomness
Emotional center
Theory: emotions as modulators of global processing
Mia: affective tonality system + emotional contour — emotions continuously modify the morphological field
Systemic loop
Theory: global feedback, regulation of the entire system
Mia: feedback and self-regulation mechanisms
Hardware — 28 head servos
Facial servos
- Eyes: 6 servos
- Eyelids: 2 servos
- Eyebrows: 4 servos
- Smile: 6 servos
- Lips: 3 servos
- Jaw: 1 servo
- Tongue: 3 servos
- Neck: 3 servos
Microcontroller
- Wi-Fi microcontroller
- Motor abstraction interface
- Real-time communication with cognitive engine
Vision
- Camera
- Dedicated detection service
- Face detection + recognition
- Distance estimation
Fabrication
- 30+ kg of 3D printed PLA
- Gears + ball bearings
- Latex skin (improving)
- CAD (~80% complete)
Technical roadmap
✓ Done
Project cognitive architecture
Cognitive agents organized in sequential engines, real-time loop. Living memory, self-regulation — everything is implemented and functional.
⚙ In progress
Head hardware — 28 servos
CAD in finalization, electronic wiring upcoming, motor interface to drive all facial axes.
→ Next
Body awareness
The robot learns its own morphology through random movements + camera feedback — expressions are discovered, not programmed.
◇ Future
LLM integration — language and culture
An LLM as an async layer: natural language interpretation, speech generation, semantic memory enrichment. Mia keeps her real-time cognition — the LLM responds when ready. Architecture combining embodied cognition + LLM.
Why Mia is unique
No existing cognitive architecture combines all these elements — not SOAR, ACT-R, LIDA, or AKOrN.
✓ Project cognition implemented and operational
✓ Embodied emotions (tonality + affective contour)
✓ Real physical embodiment (servos, camera, microcontroller)
✓ Persistent learning across sessions
✓ Morphology as cognitive control principle
✓ Living memory and self-regulation