01 Overview

ESP-Hi is a low-cost AI chat robot dog built around the ESP32-C3, targeting education and maker projects.

It captures voice via a microphone, connects to multiple large language models over Wi-Fi for conversation, and plays replies through a speaker while showing expressions on a color display.

Motion wake-up: clap or specific gestures trigger conversation without pressing buttons.

A dozen pre-programmed actions are included, such as walking, sitting, tail wagging, for quick demos and further development.

This is not a finished toy but a programmable hardware platform; users need to configure Wi-Fi and LLM API themselves.

Ideal for STEM education, AI interaction prototypes, and robotics hobbyists.

02 Hardware

• MCU: ESP32-C3, RISC-V single-core, Wi-Fi/BLE, handles audio processing, motion control, and cloud communication.
• Power: Battery-powered, boosted to 5V for servos via SX1308; logic at 3.3V. See schematic for exact voltages.
• Servos: 4 servos drive leg movements, PWM controlled; mind peak current and decoupling.
• Display: ST7789 color LCD, SPI interface for expressions and status.
• Microphone: Analog mic, sampled via ADC; light-blocking and noise-reducing structure required to lower pickup noise.
• Speaker: Audio via I2S or DAC; see schematic for interface.
• Connectors: USB for flashing/debugging, battery, servo, display FPC; see schematic.
• Assembly notes: Antenna near shell; insert left and top edges first, flex shell to seat USB. Use alignment jig for screen gluing.

03 Software flow

1. Flash ESP32-C3 firmware over USB; confirm serial log before connecting mic, display, and servos.

2. Configure Wi-Fi; put LLM API endpoint in config, keep keys on machine, not in repo.

3. Initialize peripherals: ADC mic, SPI display, PWM servos, I2S speaker.

4. Main loop: capture audio, detect wake word or motion trigger.

5. On trigger, record audio, send to LLM via Wi-Fi, get reply text.

6. Synthesize speech (or use TTS service), play via speaker.

7. Update display expression based on state, optionally execute actions.

8. Error handling: retry on network failure, low battery warning, servo timeout protection.

04 Core points

ESP32-C3 has limited resources; optimize audio processing to avoid blocking servo control during LLM requests.

Assembly order affects antenna performance and mic noise; follow instructions strictly.

Servo PWM timing must be stable; avoid conflicts with Wi-Fi interrupts.

Never commit LLM keys to firmware repo; store in config or secure storage.

Action table is key for customization; edit it to add new moves without rewriting drivers.

Different ESP32 chips have different ADC/PWM characteristics; recalibrate when porting.