Autonomous Flight Commander

Autonomous Flight Commander is a multi-agent drone swarm simulator built from scratch for the AMD Developer Hackathon. The system combines a high-performance C++17 physics engine using Spatial Grid partitioning for O(N) collision detection, PID controllers for realistic flight dynamics, and a Python-based AI orchestrator powered by Gemini 1.5 Flash that issues strategic commands to the swarm every 10 seconds via a structured JSON protocol. The architecture follows a Body–Bridge–Brain model: the C++ engine handles all physics and rendering using SFML Vertex Arrays in a single draw call; a low-latency TCP bridge streams telemetry in real time; and the Python brain processes swarm state and dispatches one of six command types — RECHARGE_CRITICAL, EVASIVE_SCATTER, FORM_GRID, MISSION_SURGE, ALTITUDE_CHANGE, or HOLD_FORMATION. The project was connected to an AMD MI300X GPU via ROCm through Antigravity, with deployment automated using a universal Makefile. A web dashboard streams the full simulation to the browser with zero client installation required. The system currently handles 10,000+ active agents at 30 FPS with sub-5ms TCP latency, and is architected for future HIP/ROCm GPU kernel offload to scale further. Use cases include autonomous logistics fleets, search and rescue coordination, and real-time 3D drone navigation.