Post by Sudais Akbar Khan

🎓 Four years of Electrical Engineering — and it all comes together in one final year project I'm incredibly proud of: Ababeel, an AI-integrated VTOL drone. The premise: commercial autonomous VTOLs are expensive, hard to repair, and lean heavily on cloud inference and proprietary telemetry. We set out to prove you can hit real capability with 3D printing and open-source hardware instead. 🛩️ Airframe A QuadPlane-configuration fixed-wing VTOL with a 1-meter wingspan, 3D-printed entirely in PLA (0.2 mm layers, 40% gyroid infill, ~168 hrs print time). Load-bearing members are reinforced with 10 mm carbon-fibre tube — modular enough that any part can be reprinted and swapped independently. 🧠 Onboard perception Real-time aerial detection runs fully on-edge on a Raspberry Pi 4 (8 GB) — no cloud, no groundlink dependency. We fine-tuned YOLOv8n on VisDrone-2019, augmented with the Stanford Drone Dataset to fix class imbalance on pedestrians/people. Final model hit 35.4% [email protected] at ~20 ms/frame, exported to NCNN (ARM NEON SIMD optimized) for 6–7 FPS on the Cortex-A72 with no accelerator. 📡 Custom telemetry Instead of costly radio modules, each drone carries an ESP32 running custom firmware, interfaced to the Pixhawk 4 over UART. It bridges the MAVLink stream onto a 2.4 GHz ESP-NOW mesh with AES encryption — a genuinely cheap, self-contained comms layer for air-to-ground telemetry and mission uploads via QGroundControl. 🎯 Flight control Pixhawk 4 on PX4 (QuadPlane VTOL), with autonomous VTOL↔cruise transition, waypoint navigation, and RTL/failsafe. Waypoint autonomy validated to ±2 m RMS. We also layered a lightweight leader-follower swarm on top — a mav.py PyMAVLink controller parsing GLOBAL_POSITION_INT broadcasts and dispatching SET_POSITION_TARGET_GLOBAL_INT to hold ~5 m formation spacing. This project stretched me across aerodynamics, embedded systems, edge ML, and flight control — and reinforced how far open-source tooling and persistence can take you. Huge thanks to my teammates and to our advisors Dr. Sadiq Amin and Dr. Salman Ghafoor for the guidance throughout. On to what's next. 🚀 #ElectricalEngineering #Drones #VTOL #ComputerVision #EdgeAI #Robotics #UAV #YOLOv8 #EmbeddedSystems #NUST #FinalYearProject #AutonomousSystems #EmbeddedSystems #PX4 #ComputerVision #YOLO #ESP32 #3DPrinting #UAV #Robotics #OpenSourceHardware

Video Content