Shashwat Batish's ALANA Is a $70 3D-Printable Humanoid Robot with LLM-Powered Conversational Smarts

Maker Shashwat Batish has released 3D print files and instructions for building ALANA, a full-scale humanoid robot — albeit one lacking legs and a head — which ties into a locally-run large language model (LLM) chatbot system for natural conversation, and costs as little as $70 in parts to put together.

"ALANA is a fully 3D-printed, general-purpose, life-size humanoid robot. The entire build cost me just around $70," Batish explains of the project. "That includes everything: 3D printing materials, electronics, and power supply. Her arms aren't just for show — they're powered by custom servo motors capable of lifting up to 500 grams [around 1.1lbs] at full extension. Not bad for a robot that costs less than the average phone charger."

The skeletal ALANA features a pair of human-like hands capable of object manipulation under autonomous control, aided by a camera system that allows for both facial and object recognition. Each arm boasts a custom mechanical design with six degrees of freedom — yet everything's under the control of a low-cost Espressif ESP8266 microcontroller, helping to bring the build costs down as low as possible.

An Espressif ESP8266 doesn't make for much of a conversationalist, though, so the heavy computational lifting is farmed out to an external computer over a Wi-Fi link. This, Batish explains, runs a local copy of the Qroq Llama 3-based large language model (LLM) for natural conversation, fed through a speech synthesis engine, plus another model for spatial awareness.

"ALANA uses 10rpm Johnson-geared motor[s] as it can provide a torque of 20kgcm and has a stall torque of 120kgcm while consuming 15W of power," Batish explains of the parts that go into the design, as well as how the overall cost was kept down. "[There is also] use of cheap structural material like PVC pipes."

The project is documented in full, with 3D print files and microcontroller firmware source code and Python scripts for running the backend system, on Instructables.