AI Game Lab

50 birds evolve simultaneously. Live neural network fires as they learn to thread pipes.

A* / Hamiltonian-cycle planner plays a perfect game — heads straight for the apple yet never traps itself. Clears the full board every run.

30 cars evolve to drive an oval track using 10 raycast sensors shown in real time.

Two DQN agents in stable self-play — fine-tuning around an analytic intercept policy, anchored so they stay sharp instead of forgetting.

100 ships evolve to dodge infinite asteroids using 10 raycast inputs. Sensor lines glow.

Monte Carlo Tree Search plays 2048 with 200 simulations per move. Live bar chart shows move scores.

DQN agent learns to break all bricks. Epsilon decay curve and Q-value bars update live.

40 prey and 4 predators evolve independent neural brains. A red-queen arms race produces emergent hunting and fleeing.

40 landers evolve to touch down on the platform. Value heatmap shows safe zones.

A 5-network ensemble trained by A* imitation + DAgger solves any random maze — 100% solve rate. Pheromone trails mark the chosen path.