Tars

We built a real-life model of TARS from Interstellar with servo-driven joints bringing sci-fi robotics to life through creative design and practical engineering

Comment

Inspiration

Our inspiration came from Interstellar. We wanted to bring the iconic robot TARS to life — not just as a replica, but as a working model that can move, talk, listen, and interact like the film version.

What it does

Moves with servo-driven joints similar to TARS’s reconfigurable walking style. Lifts using an N20 motor + lead screw system. Displays logs and system info on a 3.5″ TFT screen. Sees through an ESP32-CAM (2 MP). Listens via a MEMS microphone and speaks through speakers + amplifier. Runs AI features like speech-to-text, wake words, and GPT-based responses. Since we couldn’t get a Raspberry Pi, the AI model and core code run on a laptop, with TARS communicating via Wi-Fi to perform tasks.

How we built it

Designed the full model in FreeCAD. 3D printed the chassis and panels. Integrated servo motors, N20 motor with lead screw, ESP32-S3, ESP32-CAM, mic, speakers, amplifier, DAC, TFT display, and drivers into the body. Created wiring schematics in KiCad. Programmed the ESP32 to connect to a laptop, where speech recognition and AI processing run, then send commands back to TARS.

Challenges we ran into

Fitting components into the compact body of TARS. Not being able to afford a Raspberry Pi, which forced us to split functionality across multiple modules. Ensuring motors had enough torque to handle articulation and lifting

Accomplishments that we're proud of

Building a talking AI system that: Activates with a custom wake word. Converts speech to text using FasterWhisper. Sends queries to GPT OSS 20B via Hugging Face API. Replies with natural voice output.

What we learned

How speech-to-text systems and custom wake words work. How to connect with and use online-hosted AI models. The role of robot kinematics in determining motor size, torque, and mechanical design. Practical lessons in working with TFT displays, DACs, amplifiers, and ESP32 communication.

Components Used

Mechanical

2 × 20 kg servo motors – central articulation for each arm N20 DC motor + lead screw setup – vertical lift system 3D-printed chassis and panels – FreeCAD design Hinges, joints, structural parts

Electronics

ESP32-S3 DevKit – main controller + Wi-Fi link ESP32-CAM (2 MP) – camera vision 3.5″ SPI TFT display – info/log display INMP441 MEMS microphone – audio input PAM8406 amplifier + 2 × 3 Ω 5 W speakers – audio output UDA1334 DAC – audio interface TB6612FNG motor driver – controls N20 motor 6-cell Li-ion battery (40 A BMS) – main power 20 A buck converter (5 V) – powers motors, cam, and speakers 3.3 V buck converter – powers electronics

Software

FreeCAD – CAD design KiCad – wiring schematics FasterWhisper – transcription Custom wake-word system GPT OSS 20B via Hugging Face API – AI brain ESP32 firmware + Python scripts – control + communication

What's next for Tars

Complete the full build (this version was partial due to missing parts). Fix scaling issues (the prototype was too small). Improve the lift system for stability and smooth motion. Upgrade to better embedded hardware to remove the need for a laptop. Push TARS toward a fully autonomous, movie-like robot.

Built With

Share this project:

Updates