AEGIS-NAV

Quantum-powered emergency space navigation — real NASA/NOAA threat feeds, VQC star classification, and QAOA route optimization rendered in live 3D.

Comment

Inspiration

Space navigation in an emergency leaves no room for error. Asteroid fields, solar flares, and orbital debris move fast — and classical routing algorithms evaluate one path at a time. We wanted to show that quantum computing, even on a simulator, can evaluate entire solution spaces in superposition and make smarter routing decisions under real threat conditions.

What it does

Aegis-Nav is a real-time quantum emergency navigation system. A user triggers a mission and the system runs three quantum pipelines in sequence:

  1. Prong 1 — VQC Star Classifier: A Variational Quantum Classifier (2 qubits, ZZFeatureMap + RealAmplitudes ansatz) trained on stellar data classifies which destination is a safe harbor, returning a confidence score and quantum circuit metadata.

  2. Prong 2 — Multi-Source Hazard Mapping: Live threat feeds from NASA's NeoWs asteroid API and NOAA's DONKI solar flare database are combined with synthetic orbital debris (seeded daily for Kessler-syndrome realism) and projected onto a 100×100 threat grid.

  3. Prong 3 — QAOA Route Optimizer: A QAOA circuit evaluates all 2^6 = 64 candidate waypoint combinations simultaneously to find the lowest-cost corridor, scoring each path against the live hazard map. The quantum-optimized route is then animated on a 3D Three.js mission control display.

How we built it

  • Backend: FastAPI + Uvicorn serving five REST endpoints. All quantum circuits run locally on Qiskit's AerSimulator — no IBM Quantum account needed. The VQC trains at startup (~2 sec) so every API response is instant.
  • Frontend: React 19 + TypeScript + Vite. Three.js (via React Three Fiber) renders the 3D space grid, animated ship, color-coded hazard meshes, and both the classical and quantum paths side-by-side for comparison. Tailwind CSS v4 drives the mission-control dark UI.
  • Data pipeline: NASA and NOAA APIs are fetched on each /api/get-hazards call. If APIs are unavailable, deterministic fallbacks keep the demo live at all times.

Challenges we ran into

  • QAOA convergence: Getting QAOA to reliably pick sensible waypoints (not just the lowest-index nodes) required tuning the QUBO cost function — balancing linear hazard penalties against quadratic distance terms so the optimizer didn't collapse to trivial solutions.
  • Qiskit v1 API changes: The StatevectorSampler / Estimator primitive API changed significantly in Qiskit 1.x; adapting existing QAOA tutorials required rewriting the circuit binding and result-extraction logic from scratch.
  • 3D + React state: Keeping Three.js animation loops in sync with React state (route updates, hazard re-renders) without performance drops took careful use of useRef and useFrame.

What we learned

  • Quantum ML models (VQC) can be trained and served entirely on a local simulator in a hackathon timeframe — no cloud access required.
  • QAOA at p=1 is surprisingly capable for small routing problems and compiles in seconds.
  • Combining real-world APIs (NASA, NOAA) with quantum algorithms makes the demo feel grounded rather than purely academic.

What's next for Aegis-Nav

  • Run on real IBM Quantum hardware to compare simulator vs. hardware fidelity
  • Increase QAOA depth (p=2, p=3) and waypoint count for more complex routing scenarios
  • Add dynamic re-routing: if a new hazard appears mid-flight, trigger a QAOA re-solve in real time
  • Explore quantum error correction overlays on the 3D grid

Built With

Share this project:

Updates