AI Anti-Aircraft Turret

Fly the plane; the turret tries to shoot it down. With finite bullet speed and a bounded slew rate, the turret has to lead the target. Phase 1: playable game + an analytic intercept turret reading perfect state. Later phases add a vision model (CNN over the canvas) and a learned trajectory predictor.

Fly: Arrow keys / WASD — left·right turn, up·down throttle.

Manual turret: aim with the mouse, hold to fire.

AI turret: a pink ✕ marks the predicted intercept point; the dashed pink curve is the learned predicted path.

Predictor: Const-vel is the analytic baseline. Learned uses the ONNX model from training/ (falls back to const-vel until you train & export one).

Perception: Oracle reads the plane's true position. Vision instead regresses it from pixels with a CNN (green ring = estimate); the stats show the live tracking error vs. the oracle.

Record → Download: capture your flight as a JSON session to train the predictor on real maneuvering.