Kiwi

Kiwi is an open-source mobile voice activated AI robot that combines conversational AI with autonomous navigation.

🎥 Demo Video :

Features

🤖 Voice Assistant

• Wake-word activation using OpenWakeWord
• Natural language conversations powered by Ollama (Phi-4-mini)
• Speech-to-Text using Groq
• Text-to-Speech using Edge-TTS
• Real-time communication between the robot and backend

🚗 Autonomous Navigation

• Autonomous obstacle avoidance
• Voice-controlled movement
• Differential drive using two DC motors
• Three ultrasonic sensors for environmental awareness

---

Hardware

Voice Assistant

• ESP32 Dev Module
• INMP441 I2S Microphone
• MAX98357A I2S Audio Amplifier
• Speaker

Navigation

• Arduino Nano
• L298N Motor Driver
• 2× DC Motors
• 3× HC-SR04 Ultrasonic Sensors

---

Software Architecture

                Wake Word
             (OpenWakeWord)
                    │
                    ▼
             User Speech
                    │
                    ▼
              ESP32 Records Audio
                    │
                WebSocket
                    │
                    ▼
             Node.js Backend
        ┌─────────────────────┐
        │ Groq Speech-to-Text │
        │ Ollama (Phi-4-mini) │
        │ Edge-TTS            │
        └─────────────────────┘
                    │
                    ▼
                 ESP32
          ├── Speaker Output
          └── Motion Commands
                    │
                    ▼
              Arduino Nano
                    │
                    ▼
             Drive Motors

---

Repository Structure

firmware/
├── arduino-nano/
│   └── arduino-nano.ino
│
└── esp32/
    ├── esp32.ino
    └── config.example.h

hardware/
└── wiring_diagrams/
    ├── object_avoider.png
    └── voice_assistant.png

software/
├── Kiwi.js
├── Kiwi.py
├── package.json
└── .env.example

---

Current Capabilities

• Wake-word activated voice assistant
• Local AI conversations
• Speech recognition
• Natural speech synthesis
• Autonomous obstacle avoidance
• Voice-controlled robot movement
• Distributed architecture using ESP32, Arduino Nano and a host computer

---

In Development

• Standalone operation (no external computer required)
• More movemnet functions
• Animated eyes using a 128×64 OLED display
• Custom wake-word model
• Custom PCB

---

Getting Started

Clone the repository

git clone https://github.com/evanpgibi/Kiwi.git
cd Kiwi

Configure the ESP32

Copy:

firmware/esp32/config.example.h

to

config.h

and update the Wi-Fi credentials and backend address.

Configure the backend

Copy:

software/.env.example

to

.env

and configure the required environment variables.

Install dependencies

cd software
npm install

Start the backend

node Kiwi.js

Upload the firmware

Upload:

• firmware/esp32/esp32.ino to the ESP32
• firmware/arduino-nano/arduino-nano.ino to the Arduino Nano

---

Wiring Diagrams

The wiring diagrams for both subsystems are available in:

hardware/wiring_diagrams/

---

License

This project is licensed under the MIT License.