Resources
// CORE TECHNOLOGY · 核心技术
Everything your team needs to build an AI-powered robot. Every team gets the same foundation — what you build on top of it is entirely yours.
Hardware
ESP32-Based Boards
The ESP32 is the brain of your robot: dual-core 240 MHz processor, built-in Wi-Fi and Bluetooth, and enough compute for real-time AI interaction.
We use ESP32-based development boards (e.g. ESP32-S3, ESP32-CAM variants depending on your project).
Specifications relevant to this course: - Dual-core Xtensa LX7 @ 240 MHz - 512 KB SRAM + up to 16 MB Flash (module-dependent) - Wi-Fi 802.11 b/g/n (essential for LLM API calls) - I2C, SPI, UART, PWM, ADC — all the interfaces you’ll need
Resources: - Official ESP32 Datasheet - ESP-IDF Documentation - PlatformIO ESP32 platform
Peripherals
Teams will work with a subset of these components depending on their concept:
| Component | Purpose | Interface |
|---|---|---|
| OLED display (SSD1306, 128×64) | Text / emotion display | I2C |
| LCD display (ST7789, 240×240) | Color display | SPI |
| Microphone (INMP441) | Voice input | I2S |
| Speaker + amplifier (MAX98357) | Voice output | I2S |
| Camera (OV2640) | Visual awareness | SCCB/I2C + parallel |
| Servo / DC motor + driver | Movement, expression | PWM |
| Ultrasonic sensor (HC-SR04) | Distance / presence | GPIO |
| Capacitive touch sensor | Touch interaction | GPIO/ADC |
3D Fabrication:
Your robot’s enclosure will be 3D-modeled and printed. The physical form is part of the design — it communicates personality before the robot says a word.
Recommended modeling tools: TinkerCAD (beginner), Fusion 360 (advanced, free for students).
Frameworks
ESP-Claw
ESP-Claw is a purpose-built firmware framework for ESP32 AI robots. It abstracts the complex integration of voice, display, and LLM calls so you can focus on your idea rather than low-level plumbing.
What ESP-Claw handles for you: - Wi-Fi connection management - HTTP/HTTPS requests to LLM APIs - Audio capture and playback pipeline - Display rendering (text, icons, animations) - OTA firmware updates
Espressif engineers will give a guest lecture on ESP32 hardware and the ESP-Claw framework in Week 1.
OpenRouter
OpenRouter is a unified API gateway for large language models. Instead of integrating each model vendor separately, you write one API call and can switch between Qwen, DeepSeek, Claude, and dozens of others by changing a single parameter.
Why this matters for hardware: - Your ESP32 speaks to one endpoint — simpler networking code - Test multiple models without re-engineering your firmware - Manage costs across models in one dashboard - Automatic fallbacks if a model is unavailable
# Example: same call, different model
curl https://openrouter.ai/api/v1/chat/completions \
-H "Authorization: Bearer $OPENROUTER_KEY" \
-d '{"model": "qwen/qwen-2.5-72b-instruct", "messages": [...]}'
# Switch to DeepSeek: change one field
-d '{"model": "deepseek/deepseek-chat", "messages": [...]}'Large Language Models
Models we’ll work with
Your robot’s intelligence comes from LLMs. We’ll use models accessible via OpenRouter:
| Model | Strengths | Use when |
|---|---|---|
| Qwen 2.5 | Excellent Chinese + English, fast | Bilingual interactions, speed matters |
| DeepSeek Chat | Strong reasoning, very cost-effective | Logic-heavy interactions, budget-conscious |
| Claude (Haiku / Sonnet) | Best instruction-following, nuanced responses | Complex personality, careful tone |
Choosing a model for your robot: - Response speed matters on embedded hardware — test latency with your Wi-Fi - Cost matters if your demo runs many queries — check OpenRouter’s per-token pricing - Language quality matters for personality — do a side-by-side with your system prompt
Your system prompt and model choice are both adjustable throughout the course. Design your firmware so swapping models is a one-line change.
AI Coding Tools
Writing embedded firmware is faster with AI assistance. You will use:
Claude Code
Anthropic’s AI coding agent in your terminal. Understands your full codebase and can write, debug, and explain C++ / Arduino / ESP-IDF code.
Requires an Anthropic Pro subscription ($20/month — check if your institution has access).
npm install -g @anthropic-ai/claude-code
GitHub Copilot
In-editor AI code completion and generation inside VS Code. Excellent for boilerplate, API calls, and explaining unfamiliar code.
→ GitHub Copilot for students (free via GitHub Education)
VS Code + PlatformIO + ESP-IDF
The recommended IDE setup for ESP32 firmware development. PlatformIO handles board configuration, dependency management, and flashing. ESP-IDF is the official Espressif framework.
This setup is optional — Arduino IDE also works for getting started.
Blender
Free, open-source 3D creation suite used for sculpting and modeling organic robot enclosures. When your robot’s form needs something beyond simple geometric shapes, Blender is the right tool.
→ blender.org (free)
Version Control
All teams must maintain a GitHub repository throughout the course.
# Suggested repo structure
your-robot/
├── firmware/ # ESP32 code
│ ├── src/
│ └── platformio.ini
├── design/ # 3D files, schematics, mockups
├── research/ # User interview notes, concept brief
├── poster/ # Poster source file
└── README.md # Project description + demo photo/videoYour GitHub repo is a deliverable. Commit regularly. Write commit messages that would make sense to a stranger.