Autonomous Rescue.
High Precision. Zero Risk.

Connect human rescue teams with highly capable autonomous robots to navigate treacherous subterranean environments.

Hexy

Multimodal Search-and-Rescue Hexapod in Cave Environments

AI robot that searches dangerous caves before humans have to.

Autonomous Cave Exploration

Navigates unknown terrain without human guidance

Multimodal Survivor Detection

Visual and audio sensing for comprehensive search

Noise-Triggered Investigation

Moves toward suspicious sounds automatically

Real-Time Mapping

Generates safe paths for human rescuers

How Hexy Operates

1

Human Instructions

Chat-based command interface.

2

Autonomous Execution

Self-directed exploration and decision-making.

3

Rescue Intelligence

Safe paths delivered to teams.

Tech Stack

Back-End Stack

Grounding DINO

Survivor detection with precision 0.227 and recall 0.488

Whisper

Audio understanding for noise detection

Qwen2.5-VL

Multimodal reasoning engine

MuJoCo + Brax

Physics simulation for training

Docker + Docker Compose

Deployment orchestration

Front-End Stack

React + Typescript + Vite

Chat interface and control panel

Tailwind

Utility-first CSS styling for fast, consistent UI design

Three.js

Real-time 3D visualization

MuJoCo React

Browser-based simulation viewer

React Three Fiber

Interactive 3D rendering for immersive visualization

Ready for initialization.


terminal - hexy setup
# Initial Setup: Download Hexy core stack
$ git clone https://github.com/AMD-Hackathon-ISPM/hexy
$ cd hexy

Download Required Model

Extract the model into the root folder before proceeding with the execution methods below.

Download Model
terminal - docker
# METHOD 1: With Docker (Default)
$ docker compose up --build
terminal - local dev
# METHOD 2: Without Docker (Local Development)
# Terminal 1 (Backend):
$ cd backend
$ pip install -r requirements.txt
$ uvicorn app.main:app --reload
# serves on http://localhost:8000

# Terminal 2 (Frontend):
$ cd frontend
$ npm install
$ npm run dev
# serves on http://localhost:5173