Doc: RA1-SPEC-01
System Blueprint & Telemetry
PERCEPTION
MSS / DXCam Window Capture (1280x720 ROI)
BRAIN CORE
OpenCV matching & CUDA YOLO Nano models
EXECUTION
OS-level DirectInput hardware scan-codes
STACK
Python 3.14 + PyDirectInput
Architectural Line of Thought
PHASE 1: Deterministic Basics (Active)
Unpack MCV, parse sidebar UI, execute build loops (Power Plant, Barracks) without memory hooks.
PHASE 2: Tactical Visual Recognition
Train custom YOLO nano weights on local RTX 4060 for unit vectors and ore/gem resource nodes.
PHASE 3: Online Human Benchmark & Iteration
Deploy against online human lobbies to log live combat telemetry, measure response latency, and evolve model capability.
Milestone Progress
- Base Screen Grabber [DONE]
- Template Matching [ACTIVE]
- YOLO Object Detection [PENDING]
- Minimap Radar Parser [PENDING]
- Online Human Match vs AI [PLANNED]
Pipeline
Host: Cloudflare Pages Edge
Sync: GitHub Auto-Pipeline