Maker Lucas Fernando published the full build for a desktop companion robot on November 11 — a small, 3D-printed robot that doesn’t just display facial expressions, but tracks how much attention you give it and visibly changes mood because of it.
The personality loop is the actual project
The hardware is deliberately modest: a DFRobot FireBeetle ESP32 microcontroller, a small I2C LCD for the face, a TTP223 touch sensor standing in for petting, a passive buzzer for sound effects, and a 3.7V Li-ion battery in a custom 3D-printed enclosure. What makes it more than a display-on-a-stick is the state machine running on top: the robot cycles through neutral, happy, enjoying, excited, sleepy, sleeping, and surprised expressions depending on how recently and how often it’s been touched. Pet it regularly and its mood visibly climbs; ignore it, and it gradually gets sleepy before falling fully asleep — at which point the ESP32 drops into light-sleep mode after five minutes of no interaction, trading responsiveness for battery life. Touch it again, and it wakes back up.
Why the constraint is the interesting part
None of the individual pieces here are novel — touch sensors, small displays, and ESP32 boards are maker-shelf staples. The project’s value is in how tightly the personality system is scoped to what a $30-ish parts list can actually do: no camera, no microphone, no machine learning model, just a touch count and a timer driving seven discrete display states and a handful of buzzer melodies. It’s a clean demonstration that a believable sense of “personality” doesn’t require sensing or inferring much about the world — it requires a state machine that responds consistently to the one signal you do have.
Fully open, fully documented
The project ships under GPL3+, with a complete components list, wiring, the Arduino IDE sketch, and a note on the 3D-printed enclosure’s design choices, alongside a companion build-guide video from the maker’s own channel. It’s a reasonable intermediate-level weekend build (Hackster rates it roughly 4 hours), and a good template for anyone who wants to build their own small reactive object without reaching for a camera-and-model pipeline to get there.