Pixel Is a Walking, Voice-Controlled Desktop Robot

Desktop companion robots are everywhere right now. Scroll through any maker forum or social media feed, and you’re bound to run into at least a few of these electronic sidekicks. It’s easy to see why. They add personality and some welcome distractions to an otherwise boring workspace.

While there is no shortage of commercial options to choose from, many makers are deciding to build their own, creating custom robotic companions that are more capable and far more interesting than anything you can buy off the shelf. Hardware hacker Naz Louis, for instance, has been hard at work building his own desktop robots. Recently, he built one named Pixel that could do a lot, but, as it turned out, not enough to keep his attention for long.

The biggest problem was that Pixel couldn’t move, which made the robot feel more like a display than a companion. To fix that, Louis built a new version of Pixel that can walk. That feature alone goes a long way toward the goal of making Pixel into a better companion, but Louis didn’t stop there. He also added voice recognition to the little bot so that he could interact with it naturally.

Getting a bipedal robot to walk isn’t easy, especially when the goal is to keep the hardware simple and compact. Rather than attempting the sort of dynamic balancing seen in much larger and more powerful robots, Louis opted for a static walking approach. Pixel waddles across a desk by shifting its weight from one foot to the other before swinging a leg forward. The motion may look a little awkward, but it allows the robot to stay stable while using minimal processing power and sensor data.

The walking mechanism is driven by four analog servos. Two servos in the feet tilt the robot from side to side to shift its center of gravity, while two more in the hips rotate the body and advance each step. A dedicated servo controller handles the timing required to coordinate all four motors, freeing the main processor to focus on other tasks.

Pixel’s primary control board is a Waveshare ESP32 display module. Louis originally experimented with a smaller Seeed Studio XIAO ESP32S3 board, but it left too few available I/O pins for the features he wanted to add. The Waveshare board solved several problems at once by integrating a display, speaker, and IMU while still providing enough horsepower to run lightweight AI models locally.

Instead of relying on a cloud service or a large language model, Pixel uses a voice control system based on Espressif’s WakeNet and MultiNet models. The setup listens for a wake word and then maps spoken commands to one of roughly 200 predefined actions. Those actions include movement commands, dancing routines, timers, and a desk-guard mode. Audio feedback comes in the form of R2-D2-style beeps and chirps rather than full text-to-speech, helping keep everything running comfortably on the ESP32.

For more details on this amazing build, be sure to check out the video below.