The goal of this project is to have a animatronic robot that can engage with the user and make them feel seen and heard while conversing with an AI chatbot. The robot is self powered and and portable in a retro style case.
This project builds off of Thomas Burns' Alexatron Design. While Thomas' design used an Amazon Echo device for the speech interaction, my design will use a Raspberry Pi to run or provide real-time access to an LLM.
The CAD models for the animatronic eyes are opensource, designed by Will Cogley at Nilheim Mechatronics. The rest of the robot is designed by me and fabricated using 3D printed components and laser cut wooden panels.
Some interesting features of the robot:
8GB Ram Raspberry Pi 4B
3" CRT Display salvaged from a Panasonic TR003 portable radio TV
Speaker salvaged from a Panasonic TR003 portable radio TV
Omnidirectional USB Microphone for clear speech-to-text
Power provided using off the shelf 12V RIGID drill battery
Separated power for system isolation and troubleshooting:
CRT
Audio system
Animatronic eyes
Raspberry Pi
Travel case for portability and storage
Some initial tests were done using the OpenAI Realtime API. The interactivity was excellent, and the vocal inflections were quite impressive.
However this model has some cons:
Using the API requires internet connectivity
The API also costs money to use
Future plans:
I would like to compare latency and quality of interaction when using a small LLM running locally on the Raspberry PI vs using an API. Having a local LLM would be ideal for reliability and portability, however it will likely come at the cost of the quality of the test to speech / speech to text interaction.
One of the key components of this build is the CRT display as an audio waveform visualizer. This gives an organic, life like visual representation of the AI speaking. It was one of the most compelling parts of Thomas Burns' project that inspired me.
My project uses a tiny 3" CRT from a portable Panasonic TR003 radio television and cassette player. These were manufactured in the 1970's.
I was able to disconnect the CRT and control board from the rest of the electronics and connect an audio amp to the x axis leads of the CRT deflection coil to create the audio waveform.