Winners
SixthSense: Haptic Vision for the Blind
An affordable wearable that turns a phone's camera into directional belt buzzes, guiding blind and low-vision users around obstacles fully on-device with no internet, at a fraction of the cost of the smart canes and glasses people are priced out of.
Core trick
On-device depth and object detection on a chest-mounted phone camera map obstacles to directional belt vibrations sent over BLE to an ESP32.
Description
$2a
README excerpt
# SixthSense 🏆 - 2nd Place, Qualcomm x Meta Hackathon **An on-device navigation copilot for blind and low-vision users.** [](android/app/src/test) [](android) [](https://pytorch.org/executorch) [](android) [](firmware/esp32_belt) [](dashboard) [](LICENSE) A chest-mounted Samsung Galaxy S25 Ultra watches the path ahead, runs depth, object detection, and language models locally through ExecuTorch (with the Qualcomm QNN backend where available), folds every frame into one compact `SceneState`, and turns that state into a haptic belt buzz, a spoken answer, and a live operator view. The entire assistive path runs on the phone and keeps working in airplane mode once the app and models are on device. --- ## How It Works A single camera frame travels the whole system in tens of milliseconds, fully on the phone: 1. **CameraX** captures a frame and rotates it once into a shared upright bitmap. 2. **Depth-Anything-V2** estimates relative nearness; **YOLOv11n** detects objects. Both run as ExecuTorch `.pte` models in parallel on separate th