Taking a while?

Projects

OmniBelt

An AI navigation belt that uses computer vision to steer its wearer around obstacles, builds a spatial memory to find lost items later, and guides the hand to items with a haptic ring.

Winner — Best First-Time Hacker, KnightHacks 9Oct 2026 · 36 hours · team of four

What it does

Steers around obstacles

Computer vision on the Jetson finds obstacles in the iPhone's camera feed and maps them on a 12 × 12 m grid around the wearer. The belt plans a path through that grid and steers the wearer with its 7 vibration motors.

Finds lost items

It builds a spatial memory of what it has seen, so it can tell the wearer where an item was last seen: which direction, how far, and how long ago.

Guides the hand

A vibration motor in a finger ring guides the hand to items.

How it works

The hardware

  1. Batterypower
  2. NVIDIA Jetson Orin Nanoruns the computer-vision AI models
  3. ESP32microcontroller
  4. ULN2803motor driver
  5. 7 belt motors + 1 ring motorhaptic vibration

The software path

  1. iPhoneARKit · Apple Vision
  2. JetsonPython · FastAPI · WebSockets
  3. ESP32Arduino C++ firmware · custom serial protocol · 20 kHz PWM

Models: YOLO11n (TensorRT) with ByteTrack object tracking, Depth Anything V2 Small for depth, YOLO-World (small) and CLIP, with Gemini and ElevenLabs for voice.

lines of code
~6,200lines of code
test files
17test files
CAD / 3D-printed parts
24CAD / 3D-printed parts
haptic outputs
8haptic outputs

My part

What I did

  • Project lead and system designer.
  • Assembled and wired the physical hardware.
  • Tested, debugged and improved the integration code, and authored 22 of the team's 30 commits.
  • Gave the final pitch to the judges.

Engineering

  • Cut the computer-vision depth model's latency 3.3× (110 ms → 33 ms) by profiling it to find a library default that silently upscaled every frame.
  • Fixed memory crashes so 3 AI models share one 8 GB GPU.
  • Safety watchdog: the motors stop within 500 ms of a lost signal, and no more than 3 run at once.

Built with Andrea Gonzalez Barreto, Helena Crosby and Jayce. My teammates built the iPhone app, the vision models and the AI prompts.

DevpostGitHubBack to Projects