Capstone project exploring hybrid classical–quantum algorithms for autonomous driving and traffic perception, with an end-to-end demo stack covering planning, perception, and a remote control dashboard.
Author: Kian Haddad License: MIT
End-to-end walkthrough of the QDrive dashboard controlling the QFlow quantum traffic density pipeline on a remote device:
https://github.com/Kianmhz/QuantumDrive-Capstone/raw/main/assets/demo.mp4
If the inline player doesn't load, download the demo here.
This is a monorepo of three subprojects developed across two semesters:
| Path | Project | Stack |
|---|---|---|
semester-1-planning/ |
Quantum-Assisted Decision Making for autonomous driving — Grover's search over candidate acceleration profiles, validated in CARLA. | Python · Qiskit · CARLA |
semester-2-qflow/ |
QFlow — Quantum Traffic Density Estimation using Quantum Phase Estimation + Grover counting on YOLO occupancy grids. | Python · Qiskit · YOLOv8 · Flask |
semester-2-dashboard/ |
QDrive Dashboard — remote control panel for two device agents (Home PC + Raspberry Pi) with live MJPEG previews. | Nuxt 4 · Nuxt UI v4 · Nitro |
Each subproject has its own README.md with detailed setup, CLI reference, and architecture notes.
Formulates trajectory selection as a discrete search over candidate acceleration profiles (keep, comfort_brake, hard_brake, creep), each scored by a physics-based cost function combining safety, comfort, and tracking terms.
- Classical planner: brute-force evaluation — O(N)
- Quantum planner: Grover's search over the same candidates — O(√N)
- Result: across 120 planning snapshots, the Grover-based solver matched the classical planner's decision in 100% of cases, confirming the correctness of the quantum encoding and oracle.
See semester-1-planning/README.md for the cost function, architecture diagram, and scalability discussion.
Divides a video frame into a grid of regions, uses YOLOv8 to mark occupied cells, and applies Quantum Counting (Grover operator + QPE) to estimate density.
- Theoretical complexity: classical O(N) vs quantum O(√N) oracle queries
- Per-frame logging of classical vs quantum counts, circuit depth, transpile time, and estimated speedup
- Optional Grafana Cloud push for live dashboards
- Device-agent (
Flask) exposes/status,/start,/stop,/video_feedfor remote control
See semester-2-qflow/README.md for CLI reference, environment variables, and the academic-honesty notes on the oracle assumption.
Nuxt 4 web app that controls two remote device agents over HTTP, proxied through Nitro server routes to avoid CORS.
- Per-device Online/Offline + Running/Stopped status
- Start / Stop / Refresh per device, plus global controls
- Live MJPEG preview via each agent's
/video_feed - Timestamped event log + toast notifications
- Polls device status every 8 seconds
See semester-2-dashboard/README.md for the API contract and setup.
Each subproject is independently runnable. Common entry points:
# Semester 1 — quantum planning (CARLA simulation)
cd semester-1-planning && pip install -r requirements.txt
# Semester 2 — QFlow quantum demo (no video required)
cd semester-2-qflow && pip install -r requirements.txt
python -m src.quantum.demo
# Semester 2 — Dashboard
cd semester-2-dashboard && npm install && npm run dev- Quantum: Qiskit (Grover, QPE, Aer simulator)
- Perception: YOLOv8 (Ultralytics) with optional CUDA acceleration
- Simulation: CARLA
- Backend: Python, Flask (device agents)
- Frontend: Nuxt 4, Nuxt UI v4, Tailwind
- Observability: Grafana Cloud (optional metrics push)
Released under the MIT License.