FlyRoyale
The player called Fly is controlled by a simulation of the adult fruit-fly brain wired exactly like
the real one: 134,181 neurons and 2.7 million connections from the
FlyWire connectome. Which neurons connect, and
whether each connection excites or inhibits, comes from the real fly and never changes.
Every half-second of game time, the arena is projected onto the fly's ~7,500 photoreceptors. Activity spreads
through the wiring, and the descending & motor neurons (pink in the brain view) are read out as its choice:
wait, or play a card somewhere.
It learned only by playing against copies of itself (reinforcement learning, PPO): wins are rewarded and
losses penalised. Training adjusted how strongly each neuron responds, never the wiring. The bots never
taught it; they only measure it.
Click a card (or press 1–4), then the arena. Drag to orbit, scroll to zoom, double-click to
reset the view. B expands the brain, Space pauses, R starts a new match.
Unofficial fan project, not affiliated with Supercell. FlyWire data CC BY-NC 4.0.
Source on GitHub.