Shoggoth Mini July 14, 2025
Over the past year, robotics has been catching up with the LLM era. Pi’s π0.5 can clean unseen homes. Tesla’s Optimus can follow natural language cooking instructions. These systems are extremely impressive, but they feel stuck in a utilitarian mindset of robotic appliances. For these future robots to live with us, they must be expressive. Expressiveness communicates internal state such as intent, attention, and confidence. Beyond its functional utility as a communication channel, expressiveness makes interactions feel natural. Without it, you get the textbook uncanny valley effect.
Earlier this year, I came across Apple’s ELEGNT paper, which frames this idea rigorously through a Pixar-like lamp to show how posture and timing alone can convey intention. Around the same time, I discovered SpiRobs, a soft tentacle robot that feels oddly alive with just simple movements. One system was carefully designed to express intent while the other just moved, yet somehow felt like it had intent. That difference was interesting. I started building Shoggoth Mini as a way to explore it more directly. Not with a clear goal, but to see what would happen if I pushed embodiment into stranger territory. This post retraces that process, the happy accidents, and what I learned about building robots.
Hardware
The first challenge was creating a testbed to explore the control of SpiRobs. I started very simple: a plate to hold three motors, and a dome to lift the tentacle above them. This setup wasn’t meant to be the final design, only a platform for quick experimentation. However, halfway through 3D printing, I ran out of black filament and had to finish the dome in grey. This made it look like the dome had a mouth. When my flatmate saw it sitting on my desk, he grabbed a marker and drew some eyes. It looked good: cute, weird, slightly unsettling. I used ChatGPT to explore renders, and decided that this accident would become the form factor.
Later, I mounted stereo cameras on the dome to track the tentacle. Robot eyes are eerie. You keep expecting movement, but nothing ever happens. That prediction error focuses attention even more.
The original open-spool design relied on constant cable tension, but any slight perturbation (such as testing a buggy new policy) would make the cables leave the spool and tangle around the motor shafts. The process to fix it required untying the knot at the tip holding the cables together, and dismantling the whole robot. Adding simple spool covers eliminated most tangles and made iteration dramatically faster.
Another key step was adding a calibration script and pre-rolling extra wire length. This made it possible to:
Unroll and reroll the cables to open the robot without having to untie the tip knot, speeding up iteration dramatically
Calibrate cable tension precisely and as often as needed
... continue reading