The developer built a custom AI-powered robot for his daughter using a Raspberry Pi 5 with 8 GB of RAM. The project focuses on local processing, ensuring the robot functions without an internet connection. The interface, built with Python and Tkinter, features an animated character face that displays various expressions and lip-syncing capabilities.
For the robot's 'brain,' the developer tested several GGUF models. The Qwen3-4B-Instruct-2507 model was selected for the Raspberry Pi 5 due to its balance of quality and response time, achieving a generation speed of approximately 4.3 tokens per second. Other models like Gemma E2B were also tested for faster, shorter interactions.
The audio system utilizes Vosk for speech recognition and Piper for text-to-speech, with custom settings to achieve a specific 'anime' voice profile. Hardware components include an IMX708 autofocus camera, an MG996R servo for head movement controlled via a PCA9685 board, and a 3.5-inch Waveshare DSI touchscreen display.
The developer implemented facial tracking, allowing the robot to follow movement by coordinating eye gaze and head rotation. To manage interactions, a wake-word system was added, enabling the robot to respond to a specific name or visual contact. The project is currently in the prototype stage, with plans for further functionality.
