# 🤖 OLAF **An open-source AI companion robot. Built in public.**

▶ Watch OLAF cycle through its expressions

--- ## The Journey: 20 Weeks and Counting

Week 0
Initial concept sketch

Week 8
3D CAD design complete

Week 20
Head alive — eyes, ears & neck expressing

**An active build-in-public project.** Built while relocating cities and starting a new job — learning CAD, ROS2, embedded systems, and AI integration as I go, with AI coding assistants accelerating development. The journey continues! --- ## What is OLAF? OLAF is a **personality-first robotics framework** that brings AI agents into physical form. Think R2D2's charm meeting modern AI capabilities — a 2-3 foot tall companion that communicates through expressive OLED eyes, articulated ears (Chappie-inspired), an animated heart display, R2D2-style beeps, and LED status indicators. **This is not another voice assistant in a box.** When you BUILD your AI companion, something fundamental changes. It becomes a partner, not a servant. JARVIS challenges Tony Stark's ideas. R2D2 takes initiative. The difference? Tony BUILT JARVIS. Luke CHOSE R2D2 and they grew together. **Building creates relationship, and relationship enables true collaboration.** --- ## The Problem We're Solving AI assistants (Alexa, Siri, ChatGPT, Claude) have impressive reasoning but remain fundamentally **disembodied** — trapped in screens and speakers. Meanwhile, building physical AI with personality forces impossible choices: | Option | What You Get | What You Lose | |--------|-------------|---------------| | **Commercial robots** (Boston Dynamics, Vector) | Polish, capability | $1000s+, closed-source, no customization | | **Educational platforms** (Poppy, Niryo) | Modularity, learning | AI integration, personality, high cost | | **ROS2 robots** (Linorobot2, Andino) | Excellent navigation | Zero personality, assistant features | | **Hobbyist kits** (Otto DIY) | Affordable, charming | Limited capabilities, no AI | | **Closed AI robots** (HIWONDER) | AI + robotics bridge | Generic design, closed ecosystem | **The Core Gap:** No open-source framework integrates modular architecture + AI conversation + personality expression + practical assistance + maker-accessible approach + community ecosystem. **The Opportunity:** All the pieces are finally here — 3D printers, powerful LLMs, modern SBCs (Raspberry Pi 5), AI coding assistants, local AI accelerators (Hailo). **This is the democratization moment.** The future of embodied AI belongs to builders, not just consumers. --- ## Architecture: Three Layers, Five Modules > **Repo scope (Phase 2, 2026-05-15).** This repository is the **OLAF body / expression engine** — the Orchestration + Module layers below. The Intelligence Layer (STT, reasoning, personality, mood/emotion decisions) is a **separate sibling project, `olaf_companion`**, which publishes 4 canonical ROS 2 topics (`mood`, `activity`, `speech_emotion`, `vocalization`). This repo subscribes to those and renders them on the body. The "personality-first" vision is unchanged at the *system* level; it's just split across two repos. See `docs/sprint-change-proposal-2026-05-15.md`. ``` ┌─────────────────────────────────────────────────────────────────┐ │ INTELLIGENCE LAYER (Hybrid AI) — sibling repo olaf_companion │ │ │ │ Local AI (Hailo-accelerated): │ │ • Whisper STT (<200ms latency) │ │ • Real-time vision processing │ │ │ │ Cloud AI Agents (Claude/GPT-4): │ │ • Natural language understanding │ │ • Complex reasoning & decision-making │ │ • Tool use, function calling, multi-step planning │ │ │ └──────────────────────┬──────────────────────────────────────────┘ │ 4 canonical ROS 2 topics │ (mood · activity · speech_emotion · vocalization) ┌──────────────────────▼──────────────────────────────────────────┐ │ ORCHESTRATION LAYER ◀── THIS REPO │ │ Raspberry Pi 5 16GB + Fusion HAT+ + Hailo AI Kit │ │ Ubuntu 24.04 • ROS2 Jazzy │ │ │ │ • Expression Engine (subscribes to olaf_companion's 4 topics) │ │ • Composes activity → mood → speech_emotion → vocalization │ │ • Renders pose/LED/eye/heart from expression_map.yaml │ │ • Direct Hardware Control (servos, LEDs, kickstand via HAT) │ │ • I2C Master for ESP32 modules │ │ │ └──┬───────┬───────┬─────────────────┬───────────────────────────┘ │ │ │ │ │I2C │USB×2 │HAT WS2812 │HAT PWM │ │ │ │ ┌──▼───┐ ┌─▼─────────────┐ ┌────────▼───┐ ┌──▼──────┐ │ HEAD │ │ WAVESHARE USB │ │ INDICATOR │ │KICKSTAND│ │ESP32 │ │ Neck + Ears │ │ 24 LEDs │ │ Servos │ │(0x10)│ │ Servo Adapters│ │ (3 strips) │ │(at base)│ └──────┘ └───────────────┘ └────────────┘ └──▲──────┘ ┌──▼───┐ │ │ BASE │ │ │ESP32 │ │ │(0x11)│ │ └──────┘ ``` ### The Five Modules (MECE Principle) Each module owns its domain exclusively — no cross-dependencies. **Hybrid control:** ESP32s for real-time tasks, Pi direct control for latency-tolerant hardware. **1. HEAD Module (ESP32, I2C 0x10)** - **Hardware:** 2× GC9A01 round OLED eyes (240×240, SPI 30-60 FPS), ESP32-S3 for real-time eye animation - **Personality:** Animated eye expressions, smooth emotional transitions, blink patterns - **Why ESP32:** 60 FPS animation requires dedicated real-time control **2. BASE Module (ESP32, I2C 0x11)** - **Hardware:** Two-wheel inverted pendulum, hoverboard BLDC motors, ODrive v3.6 (UART), BNO085 9-axis AHRS (200Hz on-chip sensor fusion) - **Control:** 200Hz PID balancing loop (hard real-time, Linux can't guarantee this) - **Why ESP32:** Self-balancing requires hard real-time guarantees **3. NECK Module (Pi-controlled, USB Serial)** - **Hardware:** 3× Feetech STS3215 servos (pan/tilt/roll), Waveshare Bus Servo Adapter - **Personality:** Head orientation, expressive gestures (tilts, nods, shakes) with organic easing - **Why Pi Direct:** USB serial via Waveshare adapter — no soldering, plug-and-play **4. EARS Module (Pi-controlled, USB Serial)** - **Hardware:** 4× Feetech SCS0009 servos (2-DOF × 2 ears), Waveshare Bus Servo Adapter - **Personality:** Chappie-inspired ears — perked = alert, drooping = sad, swiveling = curious - **Why Pi Direct:** USB serial via Waveshare adapter — no soldering, plug-and-play **5. INDICATOR Module (Pi-controlled, Fusion HAT WS2812)** - **Hardware:** 3× 8-LED WS2812 strips (24 LEDs, daisy-chained) - **Functions:** interaction state (listening/working/speaking) · module health · PID visualization - **Why Pi Direct:** Single data line via Fusion HAT WS2812 port **Additional Pi-controlled hardware:** 4" heart display (animated beating heart) · kickstand (2× landing-gear servos via HAT PWM) · I2S speaker via HAT. ### Why This Architecture? - **Hybrid control topology** — ESP32s for the two real-time-critical jobs (60 FPS eyes, 200Hz balancing); Pi direct (via Fusion HAT) for everything else. Minimal soldering, fewer firmware projects, maximum reliability. - **Modular MECE** — each module completable in 1-2 weekend sprints, independently testable, contribution-friendly. - **Hybrid intelligence** — local AI (Hailo) for fast/private sensing; cloud agents for sophisticated reasoning. - **ROS2 foundation** — industry-standard pub/sub on ROS2 Jazzy / Ubuntu 24.04, with a rich ecosystem (SLAM, navigation, sensor fusion). --- ## How Expression Works The expression engine is a ROS2 node that turns "what the companion is feeling/doing" into coordinated motion + light across every channel — in real time, from a single data-driven map. ``` companion topics ─▶ activity base ─▶ + mood bias ─▶ + speech_emotion ─▶ + vocalization (mood/activity/ (posture) (LED tint / (12 authored (transient speech/voc) idle drift) emotions) gestures) │ ▼ eyes (OLED) · ears · neck · WS2812 LEDs ``` - **Single source of truth:** `ros2/src/expression_engine/config/expression_map.yaml` (+ `neck_motion.yaml`, `ears_motion.yaml`). See the [config index](ros2/src/expression_engine/config/README.md). - **Additive composition:** later layers ride on top of earlier ones, so a vocalization gesture never wipes the underlying emotional pose. - **Hand-authored, hardware-tuned:** every emotion is authored, then frozen after a pass on the real avatar. --- ## Current Status **🔄 Active build-in-public (Phase 2 — Expression Engine).** This is a work-in-progress; here's what's real today. **✅ Working on real hardware:** - **Head / eyes** — 2× OLED eyes animated by ESP32 firmware; a full set of distinct expressions (the video above) - **Ears** — 4× SCS0009 servos: perk / droop / swivel, hardware-calibrated - **Neck** — 3× STS3215 servos: pan / tilt / roll with organic easing - **Expression engine (ROS2 Jazzy)** — subscribes to the companion's 4 canonical topics and drives eyes + ears + neck + LEDs in sync - **Idle behaviour** — head drifts to sleep when idle, stirs awake on activity - **Status LEDs** — WS2812 strips reflect listening / working / speaking **🔨 In progress / next:** - Expression authoring finalization (Epic 7 — in review: speech emotions, vocalizations, activity postures, mood→LED tint) - Heart display animation (Epic 8) - Self-balancing base — PID tuning (experimental) - SLAM navigation (deferred) **🔗 Companion brain:** the AI/intelligence layer lives in the sibling repo **`olaf_companion`**, which publishes the 4 topics this body renders. **Realistic timeline:** a complex build done in spare time while working full-time. Progress is incremental, iterative, and documented as it happens. --- ## 📹 Demo Videos ### May 2026 - **OLAF expression cycling** — *2026-05-20* — the expression engine cycling through OLAF's authored emotions on real hardware (OLED eyes, articulated ears, 3-DOF neck). [▶ Watch on YouTube](https://youtube.com/shorts/9D4Txb85EnM) --- ## Repository Organization **Module-first structure.** Everything for a module lives in one place. ``` modules/ ├── head/ # OLED eyes (ESP32, I2C 0x10) ├── base/ # Self-balancing (ESP32, I2C 0x11) ├── neck/ # 3-DOF servos (Pi USB serial) ├── ears/ # Articulated ears (Pi USB serial) ├── indicator/ # WS2812 LED strips (Pi Fusion HAT) └── shared/ # Shared ESP32 libraries ``` **ESP32 modules** (head, base): `firmware/` (PlatformIO) · `hardware/` (3D models + BOM) · `tests/` · `wiring.md`. **Pi-controlled modules** (neck, ears, indicator): `hardware/` · `tests/` · `assembly.md`. **ROS2 lives separately:** ``` ros2/src/ ├── olaf_drivers/ # Driver logic (head_ears, neck, base, torso) ├── expression_engine/ # Phase 2: renders olaf_companion's 4 canonical topics └── olaf_ai/ # (deferred) STT/agents now in olaf_companion ``` **Full details:** see [repository-structure.md](docs/architecture/repository-structure.md). --- ## License & Credits **License:** Open Source (TBD: MIT or Apache 2.0) **Builder:** [Kamal Singh](https://www.linkedin.com/in/kamal-singh) - 15 years in tech (5 at Amazon) · 1 year intentional AI deep-dive · MBA + full-stack (hardware + software + AI) **Inspiration:** R2D2 (non-verbal personality) · JARVIS (AI partner, not servant) · Chappie (expressive ears) · Wall-E (friendly retro-futurism). **Built with:** AI coding assistants (Claude, GPT-4) · Raspberry Pi 5 16GB / Ubuntu 24.04 · Sunfounder Fusion HAT+ · Hailo AI Kit (26 TOPS) · ROS2 Jazzy · ESP32-S3 · Waveshare Bus Servo Adapters · OnShape · PlatformIO. --- ## Tags `#BuildInPublic` `#Robotics` `#ROS2` `#PhysicalAI` `#OpenSource` `#MakerMovement` `#AI` `#LearnInPublic` `#EmbodiedAI` `#ModularRobotics` ---

"Feel alive first, be useful second."

Star ⭐ this repo if you believe the future of AI should be open and collaborative!